Methylation and aging

Abstract

Systems and methods as described herein can determine and use methylation levels associated with various tissues and samples. For example, a method can include receiving sequence reads comprising methylation states at sites of cell-free DNA molecules. The method can further include aligning the sequence reads to one or more sets of N CpG sites or genes. Then, for each of the one or more sets of N CpG sites or genes, the method can include identifying a set of sequence reads that align to the set of one or more CpG sites or genes, and determining a methylation level using the methylation states of the set of sequence reads.

Description

[0001] Cross-references to related applications

[0002] This application claims the benefit and priority of U.S. Provisional Application No. 63 / 572,164, filed March 29, 2024, which is incorporated herein by reference in its entirety for all purposes. Background Technology

[0003] Aging generally refers to the progressive physiological changes in an organism that occur from birth over time (i.e., chronological age). These physiological changes can lead to aging, decline in biological function, and / or a reduced ability to adapt to metabolic stress. Metabolic stress can be driven by metabolic disorders influenced by environmental factors such as pathogens, temperature, noise, toxins, nutritional stress (excess or deficiency), oxidative stress, and hypoxia. Aging is a major cause of disease and disability. Chronological age is a risk factor for many diseases in the human population, such as cardiovascular disease, diabetes, cancer, Alzheimer's disease, and dementia (Partridge et al., 2018). However, the predictive power of a disease (e.g., Alzheimer's disease, cancer, cardiovascular disease, etc.) directly based on chronological age may be low (Lowsky et al., 2014). Therefore, developing improved technologies would be beneficial. Summary of the Invention

[0004] The implementation provides systems, methods, and apparatus for determining and using methylation levels associated with various tissues and samples. Examples are provided. Methylation levels can be inferred based on the methylation status of sites in cell-free DNA molecules in plasma (or other samples containing cell-free DNA, such as urine, saliva, or genital douches). Methylation levels can correspond to CpG sites or genes and can indicate the biological age of an object or a specific tissue. Therefore, one or more machine learning models can be trained to predict biological age based on the methylation levels of one or more CpG sites or one or more genes. Machine learning models can be trained using methylation level data from healthy subjects of various ages.

[0005] Furthermore, comparing a subject's biological age with its chronological age can be used to detect the presence of conditions in the subject. For example, accelerated aging in a subject or specific tissue can be detected based on this comparison. The level of accelerated aging can be used to classify the presence of conditions. When a condition is detected, it can be determined based on specific tissues exhibiting accelerated aging or based on one or more CpG sites or one or more genes that indicate accelerated aging.

[0006] In one embodiment, a method for measuring the biological age of an object is provided. The method can be executed by a computer system that receives sequence reads including the methylation status at sites on a plurality of free DNA molecules and aligns the sequence reads to a reference genome, wherein the sequence reads are aligned to N sets of one or more CpG sites. Then, for each of the N sets of one or more CpG sites, the computer system can identify a set of sequence reads aligned to the set of one or more CpG sites in the reference genome and determine the methylation level using the methylation status of the set of sequence reads. Furthermore, the computer system can: generate a feature vector based on the N methylation levels; load a machine learning model into the computer system's memory, the machine learning model being trained using training samples with reference vectors having known actual ages and measured methylation levels; input the feature vectors into the machine learning model; and use the machine learning model to predict the biological age of the object.

[0007] In another embodiment, a method for detecting lesions in an object with a known actual age is provided. The method can be executed by a computer system that receives sequence reads including methylation states at sites of multiple cell-free DNA molecules and aligns the sequence reads to a reference genome, wherein the sequence reads are aligned to N sets of one or more CpG sites. Then, for each of the N sets of one or more CpG sites, the computer system can identify a set of sequence reads aligned to the set of one or more CpG sites in the reference genome and determine the methylation level using the methylation states of the set of sequence reads. Furthermore, the computer system can: generate feature vectors based on the N methylation levels; load an age-dependent machine learning model into the computer system's memory, the age-dependent machine learning model being trained using training samples of the reference vectors having a known actual age, a known lesion classification, and measured methylation levels; input the feature vectors into the age-dependent machine learning model; and determine the classification of the presence of lesions in the object using the age-dependent machine learning model with the feature vectors.

[0008] In another embodiment, a method for detecting lesions in an object with a known actual age is provided. The method can be performed by a computer system that receives sequence reads including the methylation status at sites on a plurality of cell-free DNA molecules and aligns the sequence reads to a reference genome. Additionally, for each of a set of one or more sets of CpG sites, the computer system can: identify a set of sequence reads aligned to any CpG site in the set of CpG sites of the set, the set of CpG sites of the set comprising at least three sets of CpG sites, wherein each set of CpG sites in the set has the same morphological classification as a function of methylation level with age; and determine one or more methylation levels using the methylation status of the sequence reads of the set. Furthermore, the computer system can use an age-dependent model to determine the classification of lesions present in the object, wherein the determination uses the object's known actual age and one or more methylation levels, and wherein the model is generated using a reference sample of an object with a known classification of said lesions.

[0009] These and other embodiments of this disclosure are described in detail below. For example, other embodiments relate to systems, devices, and computer-readable media associated with the methods described herein.

[0010] A better understanding of the characteristics and advantages of the embodiments of this disclosure can be obtained by referring to the following detailed description and accompanying drawings.

[0011] Brief description of the attached figures

[0012] Figure 1 A bar chart showing the age distribution of objects according to some embodiments of this disclosure is provided.

[0013] Figure 2 A graph showing the normalized standard deviation of gene-specific 5hmC methylation levels according to some embodiments of this disclosure is displayed.

[0014] Figure 3 The figure shows a comparison between predicted age and actual age based on gene-specific 5hmC methylation levels according to some embodiments of this disclosure.

[0015] Figure 4 This is a flowchart illustrating a method for measuring the biological age of a subject according to some embodiments of the present disclosure.

[0016] Figure 5A The figure shows a comparison between age predictions based on gene-specific methylation levels and actual ages according to some embodiments of this disclosure.

[0017] Figure 5BBox plots showing accelerated aging according to some embodiments of this disclosure are displayed.

[0018] Figure 6 This is a flowchart illustrating a method for detecting diseases in a subject with a known actual age, according to some embodiments of the present disclosure.

[0019] Figure 7 The diagram shows the cfDNA 5hmC trajectories of various genes during aging, according to some embodiments of this disclosure.

[0020] Figure 8A A graph showing the linear gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0021] Figure 8B A graph showing the 5hmC trajectory of log-type gene-specific cfDNA according to an embodiment of the present disclosure is displayed.

[0022] Figure 8C A diagram showing the 5hmC trajectory of a secondary gene-specific cfDNA according to an embodiment of this disclosure is displayed.

[0023] Figure 8D A graph showing the exponential gene-specific cfDNA 5hmC trajectory according to an embodiment of this disclosure is displayed.

[0024] Figure 9A A graph showing the gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0025] Figure 9B A graph showing the gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0026] Figure 9C A graph showing the gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0027] Figure 9D A graph showing the gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0028] Figure 10A A graph showing the age-indicating gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0029] Figure 10B A graph showing the age-indicating gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0030] Figure 10CA graph showing the age-indicating gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0031] Figure 10D A graph showing the age-indicating gene-specific cfDNA 5hmC trajectory according to an embodiment of the present disclosure is displayed.

[0032] Figure 11 A graph showing the 5hmC trajectory of cfDNA during aging according to an embodiment of this disclosure is displayed.

[0033] Figure 12A The figure shows a comparison between organ-specific age predictions and actual ages according to some embodiments of this disclosure.

[0034] Figure 12B Box plots of organ-specific age acceleration according to some embodiments of this disclosure are shown.

[0035] Figure 13A The figure shows a comparison between organ-specific age predictions and actual ages according to some embodiments of this disclosure.

[0036] Figure 13B The figure shows a comparison between organ-specific age predictions and actual ages according to some embodiments of this disclosure.

[0037] Figure 13C The figure shows a comparison between organ-specific age predictions and actual ages according to some embodiments of this disclosure.

[0038] Figure 13D The figure shows a comparison between organ-specific age predictions and actual ages according to some embodiments of this disclosure.

[0039] Figure 14 A graph showing the Pearson correlation coefficient values ​​of organ-specific age models according to some embodiments of the present disclosure is displayed.

[0040] Figure 15 This is a flowchart illustrating a method for detecting diseases in a subject with a known actual age, according to some embodiments of the present disclosure.

[0041] Figure 16 The figure shows a comparison between predicted age based on cytosine methylation levels and actual age according to some embodiments of this disclosure.

[0042] Figure 17A The figure shows a comparison of the Pearson correlation coefficient between the actual age and the predicted age with the number of genetic features used in the corresponding age prediction model, according to some embodiments of this disclosure.

[0043] Figure 17B A table showing the Pearson correlation coefficient between actual age and predicted age, compared with the number of genetic features used in the corresponding age prediction model, according to some embodiments of this disclosure.

[0044] Figure 18A Another graph shows the Pearson correlation coefficient between the actual age and the predicted age according to some embodiments of this disclosure, compared with the number of CpG locus features used in the corresponding age prediction model.

[0045] Figure 18B Another table shows a comparison of the Pearson correlation coefficient between the actual age and the predicted age with the number of CpG sites used in the corresponding age prediction model, according to some embodiments of this disclosure.

[0046] Figure 19A The figures show a comparison of age predictions based on cytosine methylation levels with actual age for HCC patients at different tumor stages, according to some embodiments of the present disclosure.

[0047] Figure 19B Box plots showing the accelerated age differences in HCC patients with different tumor stages and control groups according to some embodiments of this disclosure are presented.

[0048] Figure 20 A system according to an embodiment of the present invention is shown.

[0049] Figure 21 A block diagram of an example computer system that can be used with systems and methods according to certain embodiments of the present invention is shown.

[0050] the term

[0051] “ organize "This corresponds to a group of cells that work together as a functional unit. More than one type of cell can be found in a single tissue. Different types of tissues can be composed of different types of cells (e.g., liver cells, alveolar cells, or blood cells), but can also correspond to tissues from different organisms (mother and fetus) or to tissues corresponding to healthy cells and tumor cells."

[0052] “ biological samples"Biological sample" refers to any sample obtained from an object (e.g., a human being, such as a pregnant woman, an individual with cancer, or an individual suspected of having cancer, an organ transplant recipient, or an object suspected of having a disease process involving an organ (e.g., the heart in a myocardial infarction, the brain in a stroke, or the hematopoietic system in anemia) and containing one or more nucleic acid molecules of interest. Biological samples can be bodily fluids, such as blood, plasma, serum, urine, vaginal fluid, fluid from hydrocephalus (e.g., testes), vaginal douches, pleural effusion, ascites, cerebrospinal fluid, saliva, sweat, tears, sputum, bronchoalveolar lavage fluid, peritoneal dialysis fluid, breast milk, aspirates from different parts of the body (e.g., the thyroid gland, breast), etc. Fecal samples may also be used. In various embodiments, biological samples enriched with cell-free DNA (e.g., plasma samples obtained via centrifugation protocols) may contain a majority of cell-free DNA, such as more than 50%, 60%, 70%, 80%, 90%, 95%, or 99% of the DNA. Centrifugation protocols may include, for example, 3,000 g × The liquid fraction is obtained after 10 minutes, followed by centrifugation at, for example, 30,000 g for 10 minutes to remove remaining cells. As part of the analysis of the biological sample, a statistically significant number of cell-free DNA molecules can be analyzed (e.g., to provide accurate measurement results). In some embodiments, at least 1,000 cell-free DNA molecules are analyzed. In other embodiments, at least 10,000, 50,000, 100,000, 500,000, 1,000,000, or 5,000,000 or more cell-free DNA molecules can be analyzed. At least the same number of sequence reads can be analyzed. Any quantity described herein can be any of the numbers listed above. Example sample sizes can include 30, 50, 100, 200, 300, 500, 1,000, 5,000, or 10,000 or more nanograms, or 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 ml.

[0053] the term" Comparison "", control sample "", Background Sample "", refer to "", Reference Sample "", normal "and" normal samples"This term can be used interchangeably to generally describe samples that do not have a specific disease condition or are otherwise healthy. In one example, a template-free control (NTC) sample with contaminating DNA can be considered a reference sample. In another example, a reference sample is a sample extracted from an uninfected object. Reference samples can be obtained from the object or from a database. A reference generally refers to a reference genome, which is used to map sequence readings obtained from sequencing samples from an object. A reference genome generally refers to a haploid or diploid genome that can be aligned and compared with sequence readings from a biological sample. For a haploid genome, there is only one nucleotide at each locus. For a diploid genome, heterozygous loci can be identified, such loci having two alleles, either of which allows for a match to be aligned with that locus. A reference genome (e.g., by including one or more microbial genomes) can be a reference microbial genome corresponding to a specific microbial species."

[0054] As used herein, a “fragment” (e.g., a DNA or RNA fragment) may refer to a portion of a polynucleotide or polypeptide sequence comprising at least three consecutive nucleotides. Nucleic acid fragments may retain the biological activity and / or some characteristics of the parent polypeptide. Nucleic acid fragments may be double-stranded or single-stranded, methylated or unmethylated, intact or nicked, and complexed or uncomplexed with other macromolecules (e.g., lipid particles, proteins). Nucleic acid fragments may be linear or circular. Tumor-derived nucleic acids may refer to any nucleic acid released from tumor cells, including pathogen nucleic acids from pathogens within tumor cells. As part of a biological sample analysis, a statistically significant number of fragments may be analyzed; for example, at least 1,000 fragments may be analyzed. As other examples, at least 5,000, 10,000, 50,000, 100,000, 500,000, 1,000,000, or 5,000,000 or more fragments may be analyzed, and such fragments may be selected randomly or according to one or more criteria.

[0055] The term "assay" generally refers to a technique used to determine the characteristics of nucleic acids or a sample of nucleic acids (e.g., a statistically significant number of nucleic acids) and the characteristics of the object from which the sample was obtained. An assay (e.g., a first assay or a second assay) generally refers to a technique used to determine the amount of nucleic acids in a sample, the genomic identity of the nucleic acids in the sample, the copy number variation of the nucleic acids in the sample, the methylation state of the nucleic acids in the sample, the fragment size distribution of the nucleic acids in the sample, the mutation state of the nucleic acids in the sample, or the fragmentation pattern of the nucleic acids in the sample. Any assay known to those skilled in the art can be used to detect any characteristic of the nucleic acids mentioned herein. Characteristics of nucleic acids include sequence, amount, genomic identity, copy number, methylation state at one or more nucleotide positions, size of the nucleic acid, mutations in the nucleic acid at one or more nucleotide positions, and the fragmentation pattern of the nucleic acid (e.g., the nucleotide position of the nucleic acid fragment). The term "assay" may be used interchangeably with the term "method." Assays or methods may (e.g., based on the selection of one or more cutoff values) have specific sensitivity and / or specificity, and their relative usefulness as diagnostic tools can be measured using the area under the receiver operating characteristic (ROC) curve (AUC).

[0056] The term "gene" refers to a segment of DNA involved in producing an RNA product, such as a polypeptide chain or transcription. It can include regions before and after the coding region (leader and tail) as well as insertion sequences (introns) between the individual coding segments (exons).

[0057] “ Sequence readings"Sequence reads" refers to a sequence of nucleotides obtained from any part or all of a nucleic acid molecule. For example, a sequence read can be a short nucleotide sequence (e.g., 20 to 150 nucleotides) sequenced from a nucleic acid fragment, a short nucleotide sequence at one or both ends of a nucleic acid fragment, or the sequencing of an entire nucleic acid fragment present in a biological sample. Sequence reads can be obtained in various ways, such as using sequencing technologies or using probes (e.g., in hybridization arrays or capture probes that can be used in microarrays) or amplification technologies (e.g., polymerase chain reaction (PCR) or linear or isothermal amplification using a single primer). Example sequencing technologies include high-throughput parallel sequencing, targeted sequencing, Sanger sequencing, sequencing by ligation, ion semiconductor sequencing, and single-molecule sequencing (e.g., sequencing using nanopores), or single-molecule real-time sequencing (e.g., from Pacific...). (Biosciences). This type of sequencing can be random sequencing or targeted sequencing (e.g., enriching such regions by using capture probes that hybridize with a specified region or by amplifying certain regions). Example probe-based techniques include real-time PCR and digital PCR (e.g., droplet digital PCR). As part of biological sample analysis, a statistically significant number of sequence reads can be analyzed, for example, at least 1,000 sequence reads can be analyzed. As other examples, at least 500... 0, 10,000, 50,000, 100,000, 500,000, 1,000,000, or 5,000,000 or more sequence readings. Furthermore, the number of sequence readings determined for embodiments of this disclosure can be at least 1,000, 5,000, 10,000, 50,000, 100,000, 100,000, 500,000, 1,000,000, or 5,000,000.

[0058] the term" Mapping "or" Alignment "Associating a sequence with a location or coordinates (e.g., genomic coordinates) in a reference (e.g., a reference genome) that has a known reference sequence, wherein the sequence is similar to the known reference sequence at that location in the reference. This can be done according to..." Mapping quality This is used to measure or report similarity. In one example of mapping quality used in this paper, the mapping quality X of a sequence relative to the reported position or coordinates in the reference indicates that the probability of the sequence mapping to a different position is no greater than 10^(-X / 10). For example, a mapping quality of 30 indicates that the probability of the sequence mapping to another position is less than 0.1%.

[0059] “ Reference genome "or" Reference sequence"The reference genome / sequence can be the entire genome sequence of a reference organism, one or more portions of a reference genome that may be continuous or discontinuous, a shared sequence of multiple reference organisms, a compiled sequence based on different components of different organisms, or any other suitable reference sequence. As an example, the reference genome / sequence can be at least 1,000, 10,000, 50,000, 100,000, 500,000, 10,000,000, 50,000,000, 100,000,000, 500,000,000, 1 billion, or 3 billion nucleotides long, such as the complete human genome or a repeat-masked human genome. The reference may also include information about variations of the reference known to be found in a population of organisms."

[0060] “ site (also known as " Genomic sites A "" corresponds to a single site, which can be a single base position or a group of related base positions, such as a CpG site, a TSS site, a DNase hypersensitive site, or a larger group of related base positions. A "locus" can correspond to a region that includes multiple sites. A locus can include only one site, which would make the locus equivalent to a site in the context. A region surrounding the site can be defined, such as a symmetrical or asymmetrical region around the site. As an example, a region can be included within the site. At least + / - 50 bases (e.g., 101 bases) before and after, + / - 60 bases, + / - 70 bases, + / - 80 bases, + / - 90 bases, + / - 100 bases, + / - 150 bases, + / - 200 bases, + / - 300 bases, + / - 400 bases, + / - 500 bases, + / - 600 bases, + / - 700 bases, + / - 800 bases, + / - 900 bases. And + / - 1,000 bases. As other examples, a region can be at least 100 bases, 140 bases, 147 bases, or 167 bases long. One or more regions can be analyzed, for example, to provide a lesion (e.g., cancer) grade or a proportion of a specific tissue. Various numbers of regions, sites, or loci can be analyzed, such as 50, 100, 200, 500, 1,000, 5,000, 10,000, 50,000, 100,000, 500,000, one million, or more. Various techniques, such as aligning sequence reads to a reference genome or using position-specific probes, can determine one or more genomic locations of a DNA molecule within a reference genome. Location determination can be performed on some or all of the reference genome, for example, if only a portion of the genome is analyzed. As an example, the amount of genome analyzed can be greater than 0.01%, 0.1%, 1%, 5%, 10%, or 50%. Cutting site"It can refer to the location where DNA is cut by nucleases to produce DNA fragments."

[0061] "in the mammalian genome" DNA methylation "This typically refers to the addition of a methyl group (i.e., 5-methylcytosine) to the 5' carbon of a cytosine residue in a CpG dinucleotide. DNA methylation can occur in cytosine in other backgrounds, such as CHG and CHH, where H is adenine, cytosine, or thymine. Cytosine methylation can also be in the form of 5-hydroxymethylcytosine. Non-cytosine methylation, such as N6-methyladenine, has also been reported."

[0062] "Each genomic locus (e.g., CpG site) " Methylation index "This can refer to the proportion of the number of DNA fragments showing methylation at that site (e.g., as determined by sequence readings or probes) to the total number of readings covering that site." First based state "Reading" can refer to whether a specific site is methylated at a specific location on a DNA fragment, or whether a specific site on the genome has a specific differential methylation state, such as hypermethylation or hypomethylation. "Reading" can include information obtained from a DNA fragment (e.g., the methylation state at a site). Readings can be obtained using reagents (e.g., primers or probes) that preferentially hybridize with DNA fragments with a specific methylation state. Typically, such reagents are applied after treatment with methods that differentially modify or differentially recognize DNA molecules based on their methylation state, such as bisulfite conversion, methylation-sensitive restriction enzymes, methylation-binding proteins, anti-methylcytosine antibodies, or single-molecule sequencing techniques that recognize methylcytosine and hydroxymethylcytosine.

[0063] "Region or set of sites" methylation density"CpG methylation density" can refer to the number of readings showing CpG methylation within a region (also called a bin) or a set of sites divided by the total number of readings of sites covering the region or the set of sites. A region can include one or more sites of interest, including at least 1, 2, 3, 4, 5, 10, 20, 50, 100, 200, 500, and 1,000 sites. Sites can have specific characteristics, such as being CpG sites. Therefore, the "CpG methylation density" of a region can refer to the number of readings showing CpG methylation divided by the total number of readings of CpG sites (e.g., specific CpG sites, CpG islands, or CpG sites within a larger region) covering the region. For example, the methylation density per 100 kb bin in the human genome can be determined by the total number of unconverted cytosines (corresponding to methylated cytosines) at CpG sites after bisulfite treatment relative to the total number of readings mapped to 100 kb bins. The methylation status is determined by the proportion of all CpG sites covered by the sequence reads of the kb region. This analysis can also be performed on other bin sizes, such as 500 bp, 5 kb, 10 kb, 50-kb, or 1-Mb. The region can be an entire genome or a chromosome or a portion of a chromosome (e.g., a chromosome arm). When the region includes only CpG sites, the methylation index of the CpG sites is the same as the methylation density of the region. The “ratio of methylated cytosine” can refer to the number of cytosine sites “C” that are methylated (e.g., unconverted after bisulfite conversion) relative to the total number of cytosine bases analyzed, i.e., including cytosine in the region excluding the CpG background. Methylation indices, methylation density, and the ratio of methylated cytosine are examples of “methylation level.” Other methods known to those skilled in the art, besides bisulfite conversion, can be used to detect the methylation status of DNA molecules, including (but not limited to) enzymes sensitive to methylation status (e.g., methylation-sensitive restriction enzymes), methylation-binding proteins, and single-molecule sequencing using platforms sensitive to methylation status (e.g., nanopore sequencing). (Tse et al. Proc Natl Acad Sci USA 2013; 110: 18910-18915) or by PacificBiosciences single-molecule real-time sequencing analysis (Tse et al. Proc Natl Acad Sci USA 2021; 118:e2019768118).

[0064] “ methylation level"Methylation" is an example of relative abundance, such as the relative abundance of methylated DNA molecules (e.g., at one or more specific sites) compared to other DNA molecules (e.g., all other DNA molecules or DNA molecules that are unmethylated only at one or more specific sites). The amount of other DNA molecules can act as a normalization factor. As another example, the intensity (e.g., fluorescence or electrical intensity) of methylated DNA molecules at one or more sites can be determined relative to the intensity of all DNA molecules or unmethylated DNA molecules. Relative abundance may also include intensity per volume. Methylation levels can be determined using methylation-sensing analysis, such as methylation-sensing sequencing or PCR. Exemplary methylation-sensing sequencing may include bisulfite sequencing or, for example, single-molecule techniques using nanopores.

[0065] Differentially methylated regions (DMRs) are genomic regions (e.g., sets of sites) spanning two or more biological samples and exhibiting different levels of DNA methylation. These different DNA methylation levels can be defined by specific differences in methylation indices or densities, such as (but not limited to) 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, 99%, etc. Differentially methylated sites (DMSs) can be defined in a similar manner.

[0066] As the term "as used in this article" Classification A plus sign ("+") refers to any number or other character associated with a specific characteristic of a sample. For example, the symbol "+" (or the word "positive") can indicate that a sample is classified as having a missing or amplified portion. Classification can be binary (e.g., positive or negative) or have more classification levels (e.g., a scale from 1 to 10 or 0 to 1), including probabilities. Different techniques used to determine the classification can be combined, for example by majority voting or requiring all initial / intermediate classifications to be the same (e.g., positive), to obtain the final classification from the initial or intermediate classifications used for each of the different techniques.

[0067] The term used for methylation is " Morphological classification "Morphology" can refer to how gene methylation changes with age. Different genes will increase or decrease in different ways, and such increases or decreases will also vary. A given classification can correspond to a specific pattern of change in genes at one or more methylation levels at each age. For example, for a given set of genes, the patterns (e.g., trajectories) of change in gene methylation levels over aging may be similar. Therefore, a set of genes can have the same morphological classification. Examples of morphological classifications include linear, logarithmic, quadratic (e.g., convex or concave), and exponential types.

[0068] The term "biological age" can correspond to characteristics associated with the actual functional state of an organism, where such characteristics change over time.

[0069] the term" Cutoff value "and" threshold "Cutoff value" refers to a predetermined numerical value used in the operation. For example, a cutoff size can refer to a size that fragments exceeding which will be excluded. As another example, a threshold can be a value above or below which a specific classification applies. Any of these terms can be used in any of these situations. A cutoff value or threshold can be a "reference value" or derived from a reference value that represents a specific classification or distinguishes two or more classifications. A cutoff value can be predetermined with or without reference to the characteristics of the sample or object. For example, a cutoff value can be selected based on the age or sex of the test subject. A cutoff value can be selected after and based on the output of the test data. For example, a specific cutoff value can be used when the sequencing of a sample reaches a certain depth. As another example, a characteristic value with known classifications and measurements of one or more diseases (e.g., methylation) A reference object (a level, statistical size value, or count) can be used to determine a reference level to distinguish different symptoms and / or symptom categories (e.g., whether an object has a symptom). A reference value can be chosen as a representative of a category (e.g., the mean) or a value between two clusters of the measure (e.g., chosen to obtain desired sensitivity and specificity). As another example, a reference value can be determined based on a statistical simulation of a sample. Any of these terms can be used in either of these contexts. Such reference values ​​can be determined in various ways, as a person skilled in the art will understand. For example, a measure can be determined for two different groups of objects with different known classifications, and this can be chosen. As another example, a reference value can be determined based on a statistical simulation of a sample. Specific cutoff values, thresholds, reference values, etc., can be determined based on the desired accuracy (e.g., sensitivity and specificity).

[0070] the term" Sensitivity "or" True positive rate "(TPR)" can refer to the number of true positives divided by the sum of the number of true positives and false negatives. Sensitivity can characterize the ability of a assay or method to correctly identify the proportion of a population that truly has the disease. For example, sensitivity can characterize the ability of a method to correctly identify the number of subjects in an infected population. In another example, sensitivity can characterize the ability of a method to correctly identify one or more markers indicating infection.

[0071] the term" Specificity "or" True negative rate"(TNR)" can refer to the number of true negatives divided by the sum of the number of true negatives and false positives. Specificity can characterize the ability of a assay or method to correctly identify the proportion of a population that is truly disease-free. For example, specificity can characterize the ability of a method to correctly identify the number of subjects in a population that is not infected. In another example, specificity can characterize the ability of a method to correctly identify one or more markers indicating infection.

[0072] the term" Cancer grading "Cancer grade can refer to the presence or absence of cancer (i.e., presence or absence), stage of cancer, size of the tumor, presence or absence of metastasis, total tumor burden in the body, response to treatment, and / or other measures of cancer severity (e.g., cancer recurrence). Cancer grade can be a number or other marker, such as symbols, letters, and colors. The grade can be zero. Cancer grade can also include the condition (state) before it worsened or became cancerous. Cancer grade can be used in a variety of ways. For example, screening can check for the presence of cancer in someone who previously had no known cancer. Assessment can investigate someone who has been diagnosed with cancer to monitor cancer progression over time, investigate the effectiveness of treatments, or determine prognosis. In one implementation, prognosis can be..." This indicates the probability that a patient will die from cancer, or the probability that cancer will progress after a specified period or time, or the probability or extent of cancer metastasis. Testing can mean 'screening' or it can mean checking whether someone has cancer-like characteristics (such as symptoms or other positive tests). Various types of cancer can be graded, such as carcinoma or sarcoma, melanoma, lymphoma, and leukemia, as well as cancers from various tissues, including, for example: breast, lung, liver, colon, pancreas, stomach, bone, blood, head and neck (e.g., squamous cell carcinoma of the head and neck), throat, bladder, kidney, prostate, uterus, rectum, bile duct, brain, eye, esophagus, ovary, oral cavity, nasopharynx, thyroid, urethra, testis, vagina, and pituitary gland.

[0073] "Disease grade" A symptom (also called a disease) can refer to the quantity, extent, or severity of a disease affecting an organism, where the grade can be as described above for cancer. Another example of a symptom is rejection of a transplanted organ. Other examples of symptom lesions can include autoimmune attacks (such as lupus nephritis that damages the kidneys or multiple sclerosis that damages the central nervous system), inflammatory diseases (such as hepatitis), fibrotic processes (such as cirrhosis), fatty infiltration (such as fatty liver disease), degenerative processes (such as Alzheimer's disease), and ischemic tissue damage (such as myocardial infarction or stroke). The health status of an individual can be considered as a classification without disease.

[0074] “ Machine learning models"(ML model) can refer to a software module configured to run on one or more processors to provide the characteristics of one or more samples for classification or numerical purposes. Sample data (e.g., training samples) can be used to generate an ML model to make predictions on test data. Various numbers of training samples can be used, such as at least 10, 100, 1,000, 5,000, 10,000, 50,000, or 100,000 training samples. An example is unsupervised learning models, such as Hidden Markov Models (HMMs), clustering (e.g., hierarchical clustering, k-means, mixture models, model-based clustering, density-based noise applied spatial clustering (DBSCAN), and the OPTICS algorithm), and..." Methods for learning latent variable models include Expectation-Maximization (EM), Method of Moments (MMM), and blind signal separation techniques (e.g., Principal Component Analysis, Independent Component Analysis, Nonnegative Matrix Factorization, Singular Value Decomposition) and anomaly detection (e.g., local outlier factors and isolated forests). Another example type of model is supervised learning, which can be used with embodiments of this disclosure. Example supervised learning models can include various methods and algorithms, including analytical learning, statistical models, artificial neural networks, boosting (common algorithms), backpropagation, bagging (e.g., random forests, support vector machines (SVM), support vector regression (SVR)), and Bayesian statistics. Statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, data grouping, kernel estimator, learning automata, learning classification systems, minimum information length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifier, maximum entropy classifier, conditional random fields, nearest neighbor algorithm, probabilistic approximate correct learning (PAC), ripple down rule, knowledge acquisition methods, symbolic machine learning algorithms, sub-symbolic machine learning algorithms, minimum complexity machine (MCM), random forest, classifier ensemble, ordered classification, data preprocessing, handling imbalanced datasets, statistical relation learning, or Proaftn (a multi-criteria classification algorithm). Models can include linear regression, logistic regression, deep recurrent neural networks (e.g., Long Short-Term Memory, LSTM), and hidden Markov models. Model (HMM), Linear Discriminant Analysis (LDA), k-means clustering, density-based noise applied spatial clustering (DBSCAN), random forest algorithm, support vector machine (SVM), or any model described herein.Supervised learning models can be trained in various ways using a variety of cost / loss functions that define the error from known labels (e.g., least squares and absolute difference from known classifications), as well as various optimization techniques, such as backpropagation, steepest descent, conjugate gradients, and Newton and quasi-Newton techniques.

[0075] As used in this article, " based on " refers to " At least partially based on ", and refers to a value (or result) used to determine another value, such as a value that occurs in the relationship between the input and output of a method. As used in this article, " Derivation It also refers to the relationship between the input of a method and the output of that method.

[0076] The term "about" or "approximately" may mean within an acceptable range of error for a particular value as determined by those skilled in the art, the range of error depending in part on the manner in which the value is measured or determined, i.e., limitations of the measurement system. For example, according to practice in the art, "about" may mean within one or more standard deviations. Alternatively, "about" may mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Alternatively, particularly with respect to biological systems or methods, the term "about" or "approximately" may mean within a certain order of magnitude of the value, within 5 times, or more preferably within 2 times. If a particular value is described in this application and claims, unless otherwise stated, it should be assumed that the term "about" means within an acceptable range of error for the particular value. The term "about" may have the meaning as commonly understood by those skilled in the art. The term "about" may mean ±10%. The term "about" may mean ±5%.

[0077] Where a range of values ​​is provided, it should be understood that, unless the context explicitly specifies otherwise, each intermediate value between the upper and lower limits of that range, accurate to the tenths of the lower limit unit, is also specifically disclosed. Embodiments of this disclosure cover every smaller range between any stated values ​​or intermediate values ​​within the stated range and any other stated values ​​or intermediate values ​​within the stated range. The upper and lower limits of these smaller ranges may independently include or exclude them from the range, and each range including any boundary, no boundary, or two boundaries within a smaller range is also covered by this disclosure, subject to any specifically excluded boundaries in the stated range. Where a stated range includes one or both boundaries, ranges excluding any one or both of those included boundaries are also included in this disclosure.

[0078] Standard abbreviations can be used, such as bp, base pair; kb, kilobase; pi, picoliter; s or sec, second; min, minute; h or hr, hour; aa, amino acid; nt, nucleotide; etc.

[0079] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. While any methods and materials similar to or equivalent to those described herein may be used in practice and testing of embodiments of this disclosure, some potential and exemplary methods and materials may now be described.

[0080] Detailed description

[0081] Cell-free DNA (cfDNA) can exist naturally in short fragments in various types of biological samples, such as plasma, urine, saliva, cerebrospinal fluid, pleural effusion, amniotic fluid, peritoneal fluid, and ascites. Compared to DNA contained in specific tissues, plasma or other biological samples can carry cfDNA molecules released from dead cells in various tissues. Therefore, the detection of cfDNA from biological samples can provide a minimally invasive way to obtain DNA molecules from various tissues. This enables the detection and analysis of abnormal or diseased tissues (e.g., organs). For example, cfDNA analysis can be used for non-invasive prenatal testing (Lo et al., 1997), cancer detection (Leon et al., 1977; Mandel, 1948), and organ transplant monitoring (Lo et al., 1998).

[0082] 5-Hydroxymethylcytosine (5hmC) and 5-methylcytosine (5mC) are modified forms of cytosine that can affect gene expression. 5hmC levels can be detected from cfDNA using various techniques, such as those involving bisulfite sequencing, enzymatic digestion, chemical labeling, antibody-based enrichment, liquid chromatography-tandem mass spectrometry, etc. Cytosine methylation levels (e.g., 5hmC levels, 5mC levels, or combinations thereof) can also be detected from cfDNA using various techniques, such as bisulfite sequencing, reduced representation bisulfite sequencing, microarrays, methylated DNA immunoprecipitation, nanopore sequencing, and single-molecule real-time (SMRT) sequencing. Changes in 5hmC levels or cytosine methylation levels over time can be estimated based on cfDNA from subjects of different known actual ages. Such changes in 5hmC levels or cytosine methylation levels can be correlated with biological processes such as aging and disease progression. Therefore, the 5hmC level or cytosine methylation level in cfDNA from a specific subject can indicate the state of one or more biological processes in the subject (e.g., biological age or disease progression level).

[0083] To determine the state of a biological process, methods can be developed to analyze the 5hmC level or cytosine methylation level of one or more CpG sites. Alternatively or additionally, methods can be developed to analyze the 5hmC level or cytosine methylation level of one or more genes. For example, sequence reads can be aligned to a reference genome, the sequence reads having the methylation status of one or more sites from one or more cfDNA molecules of the subject. The methylation status can indicate the presence or absence of 5hmC or 5mC modification at each site. As a result of aligning the sequence reads to the reference genome, the CpG sites corresponding to the sequence reads can be determined, and the methylation level of a set of one or more CpG sites can be deduced. In some examples, the genes corresponding to the CpG sites can also be identified, and gene-specific methylation levels can be inferred. For example, for a given set of CpG sites corresponding to a given gene, the corresponding gene-specific methylation level (e.g., gene-specific 5hmC or cytosine methylation level) can represent the number of CpG sites with 5hmC or 5mC modification relative to the total number of sites associated with said gene.

[0084] Additionally, models can be developed for predicting the state of biological processes. For example, one or more machine learning models can be developed using methylation levels (e.g., 5hmC levels or cytosine methylation levels) or combinations thereof from biological samples of subjects of different ages and disease states (e.g., healthy subjects or subjects diagnosed with a specific disease). A specific machine learning model can be trained using methylation levels from biological samples of healthy subjects of different ages. As a result of training, the machine learning model can output a predicted biological age based on inputs of methylation levels of a set of CpG sites, a set of genes, a set of CpG sites corresponding to one or more genes, or combinations thereof. Thus, the machine learning model can predict the biological age of an subject using time-dependent methylation patterns targeting a set of CpG sites, a set of genes, a set of CpG sites corresponding to one or more genes, or combinations thereof. In some examples, the predicted biological age of an subject output by the machine learning model can be compared with the subject's actual age to reveal age anomalies (e.g., accelerated or decelerated aging). Age anomalies can indicate health problems in the subject, such as the presence of a disease.

[0085] In some examples, models can be further developed for classifying the presence or progression of one or more diseases. For instance, one or more machine learning models can be trained to classify the presence or progression of a disease based on methylation levels and the known actual age of the subject. In one example, a machine learning model can be trained to classify the presence of a disease based on the difference between the known actual age and the predicted biological age. In another example, the machine learning model can be trained on the methylation levels of subjects within a specific age range and with known lesion classifications (e.g., for a disease that is positive or negative). Based on the training, the machine learning model can then output a classification of the presence or progression of a disease based on the methylation levels of biological samples from subjects within said age range.

[0086] Additionally, CpG sites, genes, or CpG sites containing corresponding genes used for age prediction, disease classification, or combinations thereof can be associated with a specific tissue type, such as a specific organ. For example, a specific tissue type may include bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis. The tissue type associated with each CpG site and / or gene can be determined based on CpG sites and / or genes enriched in biological pathways corresponding to said tissue, or based on CpG sites and / or genes highly expressed in said tissue. The methylation patterns of CpG sites and / or genes associated with a specific tissue type can then be used for tissue-specific or organ-specific age prediction or disease progression analysis. For example, machine learning models can be trained on the methylation levels of CpG sites and / or genes associated with a specific tissue to predict the biological age of an individual. As a result, acceleration or deceleration in the aging of a tissue (e.g., an organ) can be detected. In other examples, machine learning models can be trained to classify the presence or progression of disease based on the methylation levels of CpG sites and / or genes associated with a specific tissue.

[0087] As a result of obtaining methylation levels (e.g., 5hmC or cytosine methylation levels) from cfDNA and developing methods for predicting age, disease onset, or disease progression based on methylation levels, a deeper understanding of the relevant biological processes can be achieved. For example, a deeper understanding of the effects of disease on specific organs or the effects of aging can be gained. This can facilitate the development of methods for the effective detection and treatment of diseases. For instance, organ aging assessment based on methylation levels in cfDNA enables disease detection in a minimally invasive manner, which could lead to the development of new preventative interventions.

[0088] I. Introduction

[0089] Biological age reflects an organism's age based on physiological or molecular evidence. Biological age can be correlated with age-related biological processes and pathological physiological states. For example, if a subject is exceptionally healthy, their biological age may be lower than their chronological age; this can be termed "decelerating biological aging." Conversely, "accelerated biological aging" may be detected in subjects with immune-related and / or organ-related dysfunctions, and this can indicate a high risk of developing one or more diseases. Therefore, determining biological age is important for preventative diagnosis and precision medicine. A standard curve between biological age and physiological or molecular evidence can be constructed from a defined normal subject population, allowing for the quantification of biological age for each test sample. Normal subjects can be defined as those for whom no disease or symptom was detected during the study.

[0090] Recent advances in molecular biology and omics technologies have made it possible to characterize biological aging at the molecular level, and numerous omics-based aging clocks have been proposed to estimate human biological age (Rutledge et al., 2022). Based on DNA cytosine-phosphate-guanine (CpG) methylation, Hannum et al. used blood samples to predict chronological age (Hannum et al., 2013), and Horvath et al. constructed a pan-tissue methylation aging clock applicable to all human tissues (Horvath, 2013). Furthermore, blood plasma carries circulating proteins that change during aging; based on this, Lehallier et al. developed an accurate model for predicting chronological age (Lehallier et al., 2019), and Oh et al. recently demonstrated an organ-specific proteomics aging clock in living individuals (Oh et al., 2023). Transcriptomic clocks are another type, and aging clocks can be derived using transcriptomic data from different tissues, such as peripheral blood (Peters et al., 2015) and dermal fibroblasts (Fleischer et al., 2018). Furthermore, metabolomics-based aging clocks have been generated by analyzing metabolites in urine (Hertel et al., 2016) and blood (Robinson et al., 2020).

[0091] 5-Methylcytosine (5mC) is the major methylated form of DNA, and TET enzymes gradually oxidize 5mC to a series of intermediate states, such as 5-hydroxymethylcytosine (5hmC), during active demethylation (Greenberg and Bourc'his, 2019). Studies have shown that 5hmC not only acts as an intermediate in demethylation but also as a stable epigenetic marker with independent regulatory functions (Lister et al., 2013; Pastor et al., 2011). Neural development may be associated with the enrichment of 5hmC in brain and neuronal cells (Kriaucionis and Heintz, 2009; Szulwach et al., 2011), and 5hmC accumulation appears to be essential for maintaining the pluripotency of embryonic stem cells (Koh et al., 2011). Recent mouse model studies have revealed that 5hmC accumulates in a variety of aging tissues, including the liver, heart, and lungs, while 5mC showed no detectable difference between young and aging organs (Occean et al., 2023). Furthermore, Xiong et al. demonstrated a negative correlation between chronological age and 5hmC levels in human blood cells (r = -0.865), while blood 5mC showed a much weaker correlation with age (r = -0.232) (Xiong et al., 2015).

[0092] However, analysis of blood cells cannot reflect aging in other organs. Plasma DNA, which can originate from different organs, theoretically offers an opportunity to assess organ-specific aging. However, a method for this purpose based on the analysis of plasma DNA has not yet been established.

[0093] II. Methylation variability in genes across age groups

[0094] Data can be obtained from subjects of different ages, and for each gene in the gene set of each subject, this can include gene-specific methylation levels. Gene-specific methylation levels can be 5hmC levels, which can indicate the concentration or amount of 5hmC modification present in each gene of each subject. By comparing gene-specific methylation levels of subjects of different ages, highly variable genes can be identified. Highly variable genes can be those that exhibit the greatest variation across 5hmC levels across subjects of different ages.

[0095] A. Group

[0096] In some embodiments of this disclosure, data from a cohort of 785 subjects can be used to identify genes with time-varying cfDNA 5hmC signals. This data may be derived from two publicly available whole-genome 5hmC datasets. The first dataset may include data from subjects without cancer, comprising 522 healthy subjects and 263 subjects with chronic hepatitis B virus infection (HBV) or cirrhosis. The age range of subjects in the first dataset may be from 19 to 86 years. Figure 1 A bar chart 100 shows the age distribution of objects in the first dataset. For example, each bar in bar chart 100 could represent how many objects are included in each age group from 19 to 86 years old in the first dataset. The ages in bar chart 100 could be the objects' actual ages ("real ages"). The second dataset could include data on 47 objects with cancer. Objects with cancer could have glioma. The age range in the second dataset could span from 23 to 66 years old.

[0097] For each subject, there may be data on 17,000 genes. Data for each gene may include gene-specific methylation levels (e.g., the concentration or amount of 5hmC modification present in each gene for each subject). In one example, a 5hmC-Seal technique may be performed on one or more plasma cfDNA samples from each subject to determine the 5hmC level for each gene. Therefore, for each subject, a first dataset and a second dataset may include the stated genes and their corresponding 5hmC levels.

[0098] B. Identification of highly variable genes

[0099] Highly variable genes can be those exhibiting the greatest variation at the 5hmC level across groups. In some examples, the variability in the cfDNA 5hmC signal of each gene can be measured by calculating the standard deviation of the 5hmC level for each gene across groups. Alternatively, the variability in the cfDNA 5hmC signal of each gene can be measured by calculating the normalized standard deviation of the 5hmC level for each gene across groups. The normalized standard deviation can be obtained by scaling the standard deviation of the 5hmC level of each gene to the mean 5hmC level of genes falling into a given bin (Satija et al., 2015). For example, genes can be grouped into bins based on the mean 5hmC level. For each bin, there can be a corresponding mean 5hmC level. Therefore, the standard deviation of each gene can be normalized using the mean 5hmC level of its corresponding bin.

[0100] Figure 2 Figure 200 shows the normalized standard deviation of gene-specific methylation levels according to some embodiments of this disclosure. More specifically, Figure 2 The normalized standard deviation of the 5hmC level for each gene across groups is shown. Figure 2 In this model, the normalized standard deviation of each gene can be plotted against the corresponding mean 5hmC level. The gene with the highest variability at the 5hmC level can be the gene with the highest normalized standard deviation at the 5hmC level. For example, in... Figure 2 In the image, solid black dots represent the top 3,000 genes with the highest normalized standard deviation values ​​out of 17,000 genes.

[0101] In some examples, a logarithmic transformation can be performed on each 5hmC level before calculating the mean and standard deviation of the 5hmC levels of the genes. The mean 5hmC level of the genes can then be calculated using the log-transformed 5hmC values ​​of the genes for each object in the group. Additionally, the standard deviation of the 5hmC levels of the genes can be calculated using the log-transformed 5hmC values ​​of the genes. The standard deviation can be further normalized using the mean 5hmC levels corresponding to one or more bins. Then, based on the normalized standard deviation for each gene, highly variable genes in terms of 5hmC levels of cfDNA can be identified. Highly variable genes can be the top 10, top 50, top 100, top 500, top 1000, top 2000, top 3000, top 4000, top 5000 genes, etc., or the top 1%, top 2%, top 5%, top 10%, top 20%, top 30%, top 40%, top 50% of genes, etc.

[0102] Alternatively, measures of statistical dispersion may include interquartile range (IQR), range, median absolute deviation, mean absolute deviation, root mean square deviation (RMSD), distance correlation coefficient, etc. Any one or more measures of statistical dispersion can be used to identify genes with high 5hmC level variability.

[0103] III. Relationship between biological age and methylation level

[0104] The biological age of an object or a specific tissue can be estimated based on the methylation levels of a set of one or more sites (e.g., CpG sites), which may be partially or completely located within the corresponding genes. As an example, the methylation level can be the level of 5mC and / or 5hmC. Thus, in one example, the 5hmC level could be the 5hmC level of one or more genes. To correlate age with such methylation levels (e.g., 5hmC levels of genes) of a set of sites, a methylation level (e.g., 5hmC level) trajectory can be derived. Based on cfDNA samples from non-cancer subjects, a cytosine methylation level trajectory can show typical cytosine methylation levels of one or more CpG sites over time.

[0105] A. Aging clock based on cfDNA methylation

[0106] Methylation level (e.g., 5hmC and / or 5mC) trajectories can be derived using various algorithms, such as Least Absolute Shrinkage and Selection Operator (LASSO), Ridge Regression, Support Vector Machine (SVM), Analytic Learning, Artificial Neural Networks, Backpropagation, Boosting Algorithms (Meta-algorithms), Bayesian Statistics, Case-Based Reasoning, Decision Tree Learning, Inductive Logic Programming, Gaussian Process Regression, Genetic Programming, Grouping Methods for Data Processing, Kernel Estimators, Learning Automata, Learning Classifier Systems, Minimum Information Length (Decision Tree), Decision Graphs, etc.), Multilinear Subspace Learning, Naive Bayes Classifier, Maximum Entropy Classifier, Conditional Random Fields, Nearest Neighbor Algorithm, Probably Approximately Correct Learning (PAC) Algorithm, Ripple Downward Rule, Knowledge Acquisition Methods, Symbolic Machine Learning Algorithms, Sub-Symbolic Machine Learning Algorithms, Minimum Complexity Machine (MCM), Random Forest, Classifier Ensemble, Ordinal Classification, Data Preprocessing, Imbalanced Dataset Processing, Statistical Relationship Learning, or Proaftn (a multi-criteria classification algorithm).

[0107] Models used to derive methylation level trajectories, predict age based on methylation level trajectories, or combinations thereof, can utilize linear regression, logistic regression, deep recurrent neural networks (e.g., Long Short-Term Memory, LSTM), Hidden Markov Models (HMMs), Linear Discriminant Analysis (LDA), k-means clustering, density-based noise applied spatial clustering (DBSCAN), random forest algorithms, etc. Models trained using cfDNA methylation levels at one or more sites (e.g., CpG sites) to predict biological age are referred to in this paper as aging clocks.

[0108] 1. 5hmC - Aging Clock

[0109] In a specific example, a regression model (e.g., a LASSO regression model) can be developed to predict biological age based on the 5hmC values ​​of one or more genes. The model can use data from at least a subset of genes with high levels of variability in 5hmC, as identified in the cohort-based datasets described above. For example, the regression model can use data corresponding to… Figure 2 The 5hmC levels of the top 3,000 genes in the block points shown were used for training and testing.

[0110] To train and validate the regression model, data for each non-cancer subject in the cohort (e.g., 3,000 genes and their corresponding 5hmC levels) can be split into training and testing datasets. In a particular example, the data could be split into training and testing datasets at a ratio of 9:1. Training may involve fitting the regression model to the training dataset. That is, training may involve tuning the hyperparameters associated with the regression model to improve age prediction using a regression model based on the 5hmC levels of highly variable genes. In some examples, during training, the regression model may select a minimum set of genes sufficient to predict biological age. For example, the regression model may select 200 of the 3,000 highly variable genes as the minimum set of genes that can be used to predict biological age. The selected 200 genes are listed in Table 3 below.

[0111] After training, the regression model can be tested by inputting the 5hmC levels of at least 200 selected genes into the trained regression model. The 5hmC levels input into the regression model can be the 5hmC levels of cancer-free subjects in the test dataset. The trained regression model can then output a predicted age for subjects in the test dataset based on the 5hmC levels of the selected genes. The predicted age can be compared with the subject's true age to estimate the accuracy of the trained regression model.

[0112] Figure 3 A graph 300 shows a comparison between the predicted age based on gene-specific methylation levels and the actual age according to some embodiments of this disclosure. More specifically, graph 300 depicts the predicted age output by a regression model trained on a training dataset and the predicted age output by a regression model trained on a test dataset. Therefore, Figure 3 This includes the predicted age for each individual in the cohort who does not have cancer. The predicted age can be plotted against the true age of each individual in the cohort included in the first dataset. Figure 3 As shown, the age predicted by the trained regression model can correlate with the true age in the first dataset. To quantify... Figure 3 The correlation between predicted age and actual age described in the figure can be used to calculate the Pearson correlation coefficient. Figure 3 The Pearson correlation coefficient between the predicted age and the actual age described in the figure is 0.96.

[0113] 2. Aging clock based on cytosine methylation level

[0114] Age prediction can be based on 5hmC and 5mC levels. Cytosine methylation can include 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC). Bisulfite sequencing can be used to detect both types of methylated cytosine. In one example, during bisulfite treatment, unmethylated cytosine is converted to uracil, which is then read as thymidine during sequencing. In contrast, during bisulfite treatment, methylated cytosine, including 5mC and 5hmC, remains unchanged and is read as cytosine. Therefore, in bisulfite sequencing, cytosine methylation levels (e.g., signals indicating 5mC and 5hmC levels at CpG sites) can be determined by the ratio of reads identified as cytosine. In other examples, cytosine methylation patterns can be obtained using other techniques such as degenerate representative bisulfite sequencing, microarrays, methylated DNA immunoprecipitation, nanopore sequencing, and single-molecule real-time (SMRT) sequencing.

[0115] Biological age of an object or a specific tissue can be estimated based on cytosine methylation at one or more CpG sites (CpGs) and / or genes. To correlate age with cytosine methylation levels, a cytosine methylation level trajectory can be derived. Based on cfDNA samples from cancer-free subjects, a cytosine methylation level trajectory can show typical cytosine methylation levels of one or more CpGs over time.

[0116] Additionally, aging analysis can be performed on individuals with disease. Individuals with disease may exhibit abnormal results during age analysis. For example, accelerated or decelerated aging can indicate a specific disease. Therefore, by predicting age based on cytosine methylation levels and comparing the predicted age with the actual age, diseases or other lesions can be detected.

[0117] Various modeling techniques (as described above) can be used to derive cytosine methylation level trajectories, such as Least Absolute Contraction and Selection Operator (LASSO), Ridge Regression, Support Vector Machine (SVM), Analytic Learning, Artificial Neural Networks, Backpropagation, Boosting Algorithms (Meta-algorithms), Bayesian Statistics, Case-Based Reasoning, Decision Tree Learning, Inductive Logic Programming, Gaussian Process Regression, Genetic Programming, Grouping Methods for Data Processing, Kernel Estimators, Learning Automata, Learning Classifier Systems, Minimum Information Length (Decision Tree), Decision Graphs, etc., Multilinear Subspace Learning, Naive Bayes Classifiers, Maximum Entropy Classifiers, Conditional Random Fields, Nearest Neighbor Algorithms, Probably Approximately Correct Learning (PAC) Algorithms, Ripple Downward Rule, Knowledge Acquisition Methods, Symbolic Machine Learning Algorithms, Sub-Symbolic Machine Learning Algorithms, Minimum Complexity Machine (MCM), Random Forests, Classifier Ensembles, Ordinal Classification, Data Preprocessing, Imbalanced Dataset Processing, Statistical Relationship Learning, or Proaftn (a multi-criteria classification algorithm).

[0118] Other examples of models used to derive cytosine methylation level trajectories, to predict age based on cytosine methylation level trajectories, or combinations thereof, include linear regression, logistic regression, deep recurrent neural networks (e.g., Long Short-Term Memory, LSTM), hidden Markov models (HMM), linear discriminant analysis (LDA), k-means clustering, density-based noise applied spatial clustering (DBSCAN), random forest algorithms, etc. Models trained using the cytosine methylation levels of one or more CpGs and / or genes to predict biological age may be referred to in this paper as aging clocks (or aging clock models).

[0119] In a specific example, a regression model (e.g., a LASSO regression model) can be developed to predict biological age based on the cytosine methylation levels of one or more CpGs. An L1-regularization term can be incorporated into the regression model to prevent overfitting. To train and validate the regression model, data indicating the cytosine methylation levels of subjects without cancer can be obtained.

[0120] In a specific example, whole-genome bisulfite sequencing (median: approximately 193 million paired-end reads) can be performed on plasma cfDNA samples from 136 cancer-free subjects to obtain data including cytosine methylation levels from the subjects' CpG. The data can then be split into training and testing datasets. In this specific example, the 136 cancer-free subjects were split into training and testing datasets at a ratio of 4:1. Thus, the training dataset includes data indicating the cytosine methylation levels of a training subset of subjects, and the testing dataset includes data indicating the cytosine methylation levels of a test subset of subjects.

[0121] Training a regression model can include fitting the regression model to a training dataset. That is, training can include tuning hyperparameters associated with the regression model to improve age predictions using a regression model based on the cytosine methylation levels of a subset of the training data. In some examples, during training, the regression model can select genes and / or CpGs whose methylation levels predict age. Additionally, in a specific example, the regression model selects 49 CpGs, listed in Table 4 below.

[0122] After training, the regression model can be tested by inputting the cytosine methylation levels of a test subset of the objects into the trained model. The trained regression model can then output the predicted age of the test subset of the objects based on the cytosine methylation levels of the selected CpGs. The predicted age of the objects can be compared with their actual ages to estimate the accuracy of the trained regression model.

[0123] Figure 16A graph 1600 shows a comparison between predicted age based on cytosine methylation levels and actual age according to some embodiments of this disclosure. More specifically, graph 1600 depicts the comparison between the predicted age output by a trained regression model on the training dataset and the actual age on the training set (e.g., as shown at points 1602a to 1602c on graph 1600). Graph 1600 also depicts the comparison between the predicted age output by the trained regression model and the actual age on the test set (e.g., as shown at points 1604a to 1604c on graph 1600). Therefore, Figure 16 This includes the predicted age of each of 136 subjects who did not have cancer.

[0124] like Figure 16 As shown, the age predicted by the trained regression model is correlated with the actual age. To quantify... Figure 16 The correlation between predicted age and actual age shown can be calculated using the Pearson correlation coefficient. The Pearson correlation coefficient for the training dataset is 0.97, and for the test dataset it is 0.94. Therefore, Figure 1600 shows that the age predicted by the regression model trained using cfDNA cytosine methylation levels is closely related to the actual age in both the training and test datasets.

[0125] B. Exemplary methods for age prediction

[0126] Figure 4 This is a flowchart illustrating a method 400 for measuring the biological age of a subject according to some embodiments of the present disclosure.

[0127] At box 402, method 400 may include receiving sequence reads comprising the methylation status at sites on a plurality of cell-free DNA (cfDNA) molecules. cfDNA molecules may be obtained from plasma or other suitable biological samples from an object. Sequence reads may be determined using sequencing or probe-based techniques, such as using microarrays or in an amplification reaction (e.g., PCR) performed on a biological sample from the object. Sequence readings may be paired-end sequencing, single-molecule sequencing, targeted sequencing, etc. Sequence reads may include one or more sites on each cfDNA molecule, wherein the site may be a designated location or locus within the cfDNA molecule. In some examples, the site may be a specific type of site of interest. For example, the site of interest may include a CpG site or other suitable site, such as A. In such examples, sequence reads that do not include the site of interest (e.g., CpG site) may be received. Sequence reads that do not include the site of interest may be discarded and are not used in method 400.

[0128] Methylation status can indicate whether each site has been chemically modified by adding a CH3 group to form a 5mC or 5mhC group, or other types of methylation. In some examples, determining the methylation status of each site may include (1) treating multiple free DNA molecules (e.g., with bisulfite or restriction enzymes) or (2) analyzing the optical or electrical signals of multiple free DNA molecules at locations within a window including the site. Further details can be found in U.S. Publications 2021 / 0047679 and 2022 / 0328135.

[0129] At box 404, method 400 may include aligning the sequence reads with a reference genome. The reference genome may be a representative example of the DNA sequence of a species corresponding to the object. The sequence reads are aligned with the reference genome by searching for the location in the reference genome that best matches the sequence reads.

[0130] The sequence reads can be aligned to a set of N CpG sites in a reference genome. As an example, N is an integer that can be at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 30, 40, 50, 60, 70, 80, 90, or 100. Therefore, there can be at least any of these numbers of CpG sites. Examples of CpG sites that can be aligned to the sequence reads are provided in Table 4. Alternatively or additionally, the sequence reads can be aligned to genes in the reference genome. Genes can include any number of the 3,000 genes, for example, corresponding to… Figure 2 The points depicted and / or provided in Table 1. In some examples, at least some of the set of one or more CpG sites may correspond to a gene. Therefore, at least some sequence reads can be aligned to the set of one or more CpG sites within a gene. Table 4 shows examples of CpG sites corresponding to genes. Those skilled in the art will understand that any suitable alignment tool, such as Bowtie, BWA, BLAST, and FASTA, can be used.

[0131] At box 406, method 400 may include identifying a set of sequence reads aligned to a set of one or more CpG sites in a reference genome for each of N sets of one or more CpG sites. For example, the reference genome may include annotations or other suitable indications of the location and boundaries of CpG sites, genes, or combinations thereof. Thus, based on the alignment of the sequence reads to the reference genome, each sequence read may be associated with a CpG site, gene, or combination thereof. Each sequence read may correspond to one or more CpG sites, genes, or combinations thereof. Each of the CpG sites, genes, or combinations thereof may be associated with a specific tissue type, such as a specific organ. In one example, the specific tissue type may be selected from the group including bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis.

[0132] The association of each of the CpG sites, genes, or combinations thereof with a specific tissue type can be determined based on (1) biological pathways, (2) epigenetic patterns, or (3) expression levels above a threshold in the specific tissue type. Regarding biological pathways, a specific tissue type can be determined based on the CpG sites, genes, or combinations thereof involved in biological pathways associated with that tissue. A biological pathway can be a series of molecular events within a biological environment (e.g., a cell) that can be carried out by gene products (e.g., proteins). For example, the gene encoding insulin... INS Genes can be associated with the pancreas based on insulin, which is used in biological pathways for glucose homeostasis. Regarding expression levels, some CpG sites or genes can be expressed at high levels (e.g., above a threshold) in specific tissues. Therefore, for a given gene, if the corresponding expression level in a specific tissue (e.g., the heart) is above a threshold, then that gene can be associated with that specific tissue.

[0133] At box 408, method 400 may include determining the methylation level for each of N sets of sequence reads using the methylation status of one or more CpG sites. In some examples, the methylation level may include a count (number) of methylated sites (or unmethylated sites) clustered on a sequence read of a given CpG site or gene. For example, to determine the methylation level in the set of one or more CpG sites, the amount (number) of methylation status at one or more CpG sites indicating the presence or absence of methylation may be determined. Thus, in some examples, the methylation level of a set of sequence reads aligned to a set of one or more CpG sites may be the amount of methylation status indicating the presence of methylation at each CpG site in that set.

[0134] Alternatively, in an example where sequence reads are compared to genes in a reference genome, the methylation level can be the count (number) of methylated sites (or unmethylated sites) clustered on the sequence reads of each gene. Therefore, the methylation level can be a gene-specific methylation level. Additionally, in examples where at least some of the sets of one or more CpG sites correspond to genes, the methylation level can be a gene-specific methylation level. For example, the methylation level can be the amount of methylation state indicating the presence of methylation at each CpG site corresponding to a gene.

[0135] The number of methylation sites can be normalized. For example, the methylation level can be a proportion, percentage, or other suitable quantification of the methylation density of a set of one or more CpG sites or a set of genes. That is, the methylation level can be a measure of the number of methylation sites in a sequence read associated with a set of one or more CpG sites or a set of genes relative to the total number of sites in a sequence read associated with a set of one or more CpG sites or a set of genes.

[0136] At box 410, method 400 may include generating a feature vector based on N methylation levels. The feature vector may include the methylation levels for each set of one or more CpG sites or genes. The feature vector may include methylation levels in a structured form that can be ingested (inputted) into a machine learning model and understood by the machine learning model.

[0137] At box 412, the process may include loading a machine learning model into the memory of a computer system. The machine learning model can be trained using training samples with known actual ages (i.e., taken from training subjects with known actual ages) and a reference vector of methylation levels measured from the training samples (i.e., biological samples from the training subjects). The machine learning model may be a regression model or other suitable type of machine learning model. In one example, the training samples may be obtained from a training group (e.g., a group as described herein). The training samples may be subjects without cancer. The training group may include the known actual age of each subject without cancer. Additionally, the reference vector of methylation levels may include the 5hmC levels of at least a subset of highly variable genes.

[0138] At box 414, the process may include feeding feature vectors into a machine learning model. For example, feature vectors may be fed into a regression model. Feature vectors (and other feature vectors described herein) may have various formats, such as being organized sequentially into a single array or in a multidimensional form when other values ​​besides methylation levels are also included in the feature vectors.

[0139] At box 416, the process may include using a machine learning model to predict the biological age of an object. As an example, the predicted biological age may be an age (e.g., 20, 30, 45, 55, etc.), or a range of ages (e.g., 20 to 25, 30 to 39, etc.), or even a higher precision than age, such as a month within a range of months. In examples where genes are associated with a specific tissue type, the biological age may be an age specific to that tissue type. Additionally, in some examples, the biological age predicted using a machine learning model may be compared to the object's true actual age. If the predicted age deviates from the true actual age, for example, by a threshold amount, it can be determined that the object has a lesion (e.g., a symptom, disease, or condition). In such cases, an alarm for accelerated aging or other appropriate indicator may be generated and output. In examples where a set of one or more CpG sites or a set of genes is associated with a specific tissue type, a predicted age that is at least a threshold amount larger than the actual age may indicate the presence of a lesion affecting that tissue type.

[0140] IV. Lesion Detection

[0141] Aging analysis can be performed on diseased subjects. Diseased subjects may exhibit anomalous results during age analysis. For example, accelerated aging of specific tissues can indicate a specific disease. Therefore, diseases or other lesions can be detected by predicting age based on cfDNA methylation levels (e.g., 5hmC and / or 5mC levels) and comparing the predicted age with the subject's true age. Alternatively or additionally, lesions can be detected by analyzing the methylation levels of a set of CpG sites with highly variable methylation levels over time and / or genes, and comparing these methylation levels with reference methylation levels for a known age. The comparison can be performed using a machine learning model trained with reference methylation levels (reference vectors).

[0142] A. Example model based on 5hmC level

[0143] To study biological aging in diseased subjects based on cfDNA methylation level signals, a second dataset can be input into the regression model. As mentioned above, based on data from cancer-free subjects in the cohort, the regression model selects 200 genes that can be used to predict biological age. Of these 200 genes, 175 have detectable 5hmC levels in the second dataset. Therefore, to accommodate the differences in the datasets, the regression model can be retrained. For example, data from cancer-free subjects in the first dataset can be used to retrain the model. The cancer-free subject data used for retraining the model could include the 175 genes and their corresponding 5hmC levels. Similar to the previous example, the data can be split into training and testing datasets at a 9:1 ratio.

[0144] After training, the 5hmC levels from the test dataset can be input into the regression model. Alternatively, the 5hmC levels from the second dataset can be input into the regression model.

[0145] Figure 5A A graph 500a is shown comparing age predictions based on gene-specific methylation levels with actual ages according to some embodiments of this disclosure. Specifically, graph 500a may include predicted ages from model outputs of a test dataset and a second dataset. Figure 5A The regression model described can accurately predict the true age of subjects in the test dataset who do not have cancer. For example, the Pearson correlation coefficient between the predicted age and the true age for the test dataset is 0.94. Furthermore, as... Figure 5A As shown, the predicted age of objects in the second dataset can have a certain degree of correlation with their actual age. For example, the Pearson coefficient between the predicted age and the actual age for the second dataset is 0.66. Furthermore, as depicted, the predicted age corresponding to the second dataset can often be greater than the object's actual age.

[0146] Since the predicted age output by the regression model from the second dataset is generally greater than the actual age, it can be predicted that subjects with gliomas may exhibit accelerated aging. This prediction can be based on the gene-specific methylation levels (e.g., 5hmC levels) of subjects in the second dataset compared to control subjects (e.g., subjects without cancer in the first dataset). Accelerated aging can be defined as the difference between the predicted age and the actual age. To determine whether subjects with gliomas exhibit accelerated aging, it can be determined whether the difference in accelerated aging between control subjects and subjects with gliomas is significant. For this purpose, the Wilcoxon rank-sum test can be performed.

[0147] Figure 5B Box plot 500b shows the acceleration of age according to some embodiments of this disclosure. Specifically, box plot 500b can depict the results of the Wilcoxon rank-sum test. For example, as shown in box plot 500b, the median age acceleration in glioma subjects might be approximately 13 years, while the median age acceleration in control subjects might be approximately 1 year. The p-value of the Wilcoxon rank-sum test can be less than 0.001, indicating a significant difference between the predicted age of control subjects and subjects with glioma. Based on these results, it can be determined that glioma patients may indeed exhibit accelerated age. Previous studies (Liao et al., 2018; Lin and Wagner, 2015) have also shown that glioma tumor tissue exhibits accelerated age compared to healthy tissue based on the aging clock of Horvath methylation.

[0148] Therefore, predicting age based on gene-specific methylation levels can be used to identify the biological age of abnormally accelerated or decelerated processes. This can effectively detect, prevent, and treat diseases.

[0149] B. Combined use of 5hmc and 5mc to detect lesions

[0150] To study biological aging in diseased subjects based on cytosine methylation levels (e.g., based on 5hmC and 5mC levels), data including methylation levels obtained through whole-genome bisulfite sequencing of plasma cfDNA samples from 38 subjects with early (n = 20), intermediate (n = 11), and late (n = 7) cancers can be input into the study described above. Figure 16 In the aforementioned regression model, the regression model can then output the predicted age of individuals with early, intermediate, and late-stage cancer.

[0151] Figure 19A A graph 1900 shows age predictions based on cytosine methylation levels versus actual age for different stages of hepatocellular carcinoma (HCC) according to some embodiments of this disclosure. Graph 1900 plots the actual age of the comparison subjects (e.g., as shown by points 1902a to 1902c) against the age predictions of subjects without cancer (referred to as controls in graph 1900). Graph 1900 also plots the age predictions of subjects with early-stage cancer (e.g., as shown by point 1904), intermediate-stage cancer (e.g., as shown by point 1906), and late-stage cancer (e.g., as shown by point 1908).

[0152] Therefore, graph 1900 shows the difference between predicted and actual ages for subjects without cancer and for subjects with early, intermediate, and late-stage cancer. Subjects with early, intermediate, and late-stage cancer have larger errors (e.g., mean absolute error (MAE)) between their actual and predicted ages compared to subjects without cancer. Specifically, the MAE is 2.8 for subjects without cancer, 10.1 for those with early-stage cancer, 11.7 for those with intermediate-stage cancer, and 45.5 for those with late-stage cancer.

[0153] Figure 19BBox plot 1902 illustrates the age acceleration in subjects without cancer and subjects with HCC according to some embodiments of this disclosure. Specifically, box plot 1902 depicts the results of the Wilcoxon rank-sum test. For example, as depicted in box plot 1902, the median age acceleration in subjects with HCC could be approximately -10 years. The p-value of the Wilcoxon rank-sum test could be less than 0.001, indicating a significant difference between the predicted ages of subjects without cancer and those with HCC. Based on these results (e.g., based on a negative age acceleration and a significant difference), it can be determined that subjects with HCC may exhibit a slower age acceleration relative to subjects without cancer.

[0154] Therefore, age prediction based on methylation levels (e.g., based on 5mC and 5hmC levels) can be used to identify the biological age of abnormal accelerations or decelerations. This enables the effective detection, prevention, and treatment of diseases.

[0155] C. Other models

[0156] In another example, to study biological aging in diseased subjects based on cfDNA 5hmC signaling, a machine learning model can be trained on data from a second dataset corresponding to a subset of cancers and data from a first dataset corresponding to subjects without cancer. The data can be split into training and testing datasets. The data may include lesion classifications (e.g., glioma-positive or glioma-negative) and 5hmC levels of highly variable genes. The machine learning model can use logistic regression, one or more decision trees, neural networks, etc. During training, the machine learning model can learn to predict the classification of disease presence (e.g., whether an object has glioma) based on the 5hmC levels of a gene set. After training, the 5hmC levels from the testing dataset can be fed into the machine learning model. The machine learning model can then output a predicted classification of glioma presence for each object in the testing dataset.

[0157] In another example, to study biological aging in diseased subjects based on cfDNA 5hmC signaling, one or more machine learning models can be trained on data from a second dataset of subjects within a specific age range and data from a first dataset of subjects within the same age range. For each age range, the corresponding data can be split into training and testing datasets. The data may include each subject's age, lesion classification (e.g., glioma-positive or glioma-negative), and the level of the highly variable gene 5hmC. Machine learning models may include logistic regression, decision trees, neural networks, etc. During training, each machine learning model can learn to predict the classification of the presence of a disease or other lesion (e.g., whether the subject has glioma) based on the 5hmC level of the gene set and the subject's age. After training, similarly, the machine learning model can output a predicted classification of the presence of glioma for each subject in the testing dataset.

[0158] In another example, to study biological aging in diseased subjects based on cfDNA methylation levels, a machine learning model can be trained on data including methylation levels obtained through whole-genome bisulfite sequencing of plasma cfDNA samples from 38 subjects with early (n = 20), intermediate (n = 11), and late (n = 7) cancers, and data obtained through whole-genome bisulfite sequencing of plasma cfDNA samples from 136 subjects without cancer. The data can be split into training and testing datasets. The data may include methylation levels of lesion classifications (e.g., HCC positive or HCC negative) and sets of CpG sites (e.g., CpG sites provided in Table 4). The machine learning model can use logistic regression, one or more decision trees, neural networks, etc. During training, the machine learning model can learn to predict the classification of lesion presence (e.g., whether an object has HCC) based on methylation levels of one or more sets of CpG sites. After training, methylation levels from the testing dataset can be fed into the machine learning model. The machine learning model can then output a predicted classification of cancer (e.g., HCC) for each object in the testing dataset.

[0159] In another example, to study biological aging in diseased subjects based on cfDNA methylation levels, one or more machine learning models can be trained on data including methylation levels obtained through whole-genome bisulfite sequencing of plasma cfDNA samples from 38 subjects with early-stage (n = 20), intermediate-stage (n = 11), and late-stage (n = 7) cancers, and data obtained through whole-genome bisulfite sequencing of plasma cfDNA samples from 136 subjects without cancer. For each of the multiple age ranges of the subjects, the corresponding data can be split into training and testing datasets. The data may include each subject's age, lesion classification (e.g., HCC positive or HCC negative), and methylation levels of a set of CpG sites. Machine learning models may include logistic regression, decision trees, neural networks, etc. During training, each machine learning model can learn to predict the classification of the presence of a disease or other lesion (e.g., whether the subject has glioma) based on the methylation levels of one or more sets of CpG sites and the subject's age. After training, similar to the above, the machine learning model can output a predicted classification of the presence of cancer (e.g., HCC) for each subject in the testing dataset.

[0160] D. Lesion prediction methods

[0161] Figure 6 A flowchart illustrating an example of method 600 for detecting ailments in objects with a known actual age is shown. Some aspects of method 600 can be performed in a similar manner to method 400.

[0162] At box 602, method 600 may include receiving sequence reads comprising the methylation state at sites on a plurality of cell-free DNA (cfDNA) molecules. The sequence reads may include one or more sites on each cfDNA molecule. Box 602 may be performed in a manner similar to box 402 of method 400.

[0163] At box 604, method 600 may include aligning a sequence reading with a reference genome. The reference genome may be a representative example of a DNA sequence corresponding to a species of the object. The alignment may include searching for the location in the reference genome that best matches the sequence reading.

[0164] The sequence reads can be aligned to a set of N CpG sites in a reference genome. As an example, N is an integer that can be at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 30, 40, 50, 60, 70, 80, 90, or 100. Therefore, there can be at least any of these numbers of CpG sites. Examples of CpG sites that can be aligned to the sequence reads are provided in Table 4. Alternatively or additionally, the sequence reads can be aligned to genes in the reference genome. Genes may include those corresponding to… Figure 2 The blocky points depicted are any number of the 3,000 genes provided in Table 1. In some examples, at least some of the set of one or more CpG sites may correspond to a gene. Therefore, at least some sequence reads can be aligned to the set of one or more CpG sites within a gene. Table 4 shows examples of CpG sites corresponding to genes. Box 604 can be performed in a manner similar to box 404 of method 400.

[0165] At box 606, method 600 may include identifying a set of sequence reads aligned to each of N sets of one or more CpG sites in a reference genome. For example, the reference genome may include annotations or other suitable indications of the locations and boundaries of CpG sites, genes, or combinations thereof. Thus, based on the alignment of the sequence reads to the reference genome, each sequence read may be associated with a CpG site, gene, or combination thereof. For example, if the reads are long enough, each sequence read may correspond to one or more CpG sites, genes, or combinations thereof. Box 606 may be performed in a manner similar to box 406 of method 400.

[0166] Each of the CpG sites, genes, or combinations thereof can be associated with a specific tissue type, such as a specific organ. In one example, a specific tissue type could be selected from a group including bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis. The association of each gene with a specific tissue type can be determined based on (1) a biological pathway, (2) an epigenetic pattern, or (3) an expression level above a threshold in the specific tissue type.

[0167] At box 608, method 600 may include determining the methylation level for each of N sets of sequence readings using the methylation status of one or more CpG sites. Box 608 may be performed in a manner similar to box 408 of method 400.

[0168] For example, to determine the methylation level (e.g., 5hmC level and / or 5mC level) in a set of one or more CpG sites, the amount (number) of methylation states at the set of one or more CpG sites indicating the presence or absence of methylation can be determined. Thus, in some examples, the methylation level of a set of sequence reads aligned to a set of one or more CpG sites can be the amount of methylation states indicating the presence of methylation at each CpG site in that set. Alternatively, in examples where sequence reads are aligned to genes in a reference genome, the methylation level can be the count (number) of methylated sites (or unmethylated sites) clustered on the sequence reads of each gene. Thus, the methylation level can be a gene-specific methylation level. Additionally, in examples where at least some of the sets of one or more CpG sites correspond to genes, the methylation level can be a gene-specific methylation level. For example, the methylation level can be the amount of methylation states indicating the presence of methylation at each CpG site corresponding to a gene.

[0169] At box 610, method 600 may include generating a feature vector based on N methylation levels. The feature vector may include the methylation level of each of one or more CpG sites or the methylation levels of one or more genes. The feature vector may include methylation levels in a structured form that can be ingested and understood by a machine learning model. Box 610 may be performed in a similar manner to box 410 of method 400.

[0170] At box 612, method 600 may include loading an age-dependent machine learning model into the memory of a computer system. Various types and architectures of age-dependent machine learning models can be used. An age-dependent machine learning model may include multiple sub-models, each corresponding to a different biological age or age range.

[0171] At box 614, method 600 may include inputting feature vectors into an age-dependent machine learning model. Feature vectors (and other feature vectors described herein) may have various formats, such as being organized sequentially into a single array or in a multidimensional form when other values ​​besides methylation levels are also included in the feature vectors.

[0172] At box 616, method 600 may include using an age-dependent machine learning model to predict the classification of the lesion level (e.g., the presence of lesions, such as symptoms) in an object.

[0173] In some examples, determining the classification of lesions present in an object may include comparing a feature vector with a representative reference vector, which is determined using a set of measurement reference vectors that have the same known lesion classification (e.g., disease classification). The distance between the two vectors can be determined. If the feature vector is sufficiently similar to a first representative reference vector (e.g., within a threshold distance or the closest representative reference vector), the object can be determined to have a first classification corresponding to the first representative reference vector.

[0174] Age-dependent machine learning models can generate a first representative reference vector based on reference vectors of sample subjects who have a lesion (e.g., glioma) and are the same age as or similar to the subject (e.g., within the same age range). Therefore, the first representative reference vector may include measured (including statistically determined, e.g., average) methylation levels of CpG sites or genes of sample subjects suffering from said lesion.

[0175] Additionally, age-dependent machine learning models can generate a second reference vector for healthy sample objects of the same or similar age as the object (e.g., objects without lesions). Therefore, classifying the presence of lesions in an object can include comparing the feature vector with each representative reference vector and determining which representative reference vector the feature vector is most similar to. In some examples, there can be any number of representative reference vectors corresponding to any number of lesions.

[0176] Each actual age or age range can have its own set of representative feature vectors. Therefore, the model can include sub-models for different ages, each with a set of representative feature vectors for a different classification. In other implementations, the model can determine the distances of all ages relative to all other representative reference vectors at once, but the distances of different ages relative to such reference vectors can be weighted by scaling their dimensions, such that the distances relative to reference vectors for different ages are generally higher than the distances relative to reference vectors for the same age.

[0177] Additionally, in some examples, determining the classification of a symptom may include using an age-dependent machine learning model to predict the subject's biological age, comparing the biological age with a known chronological age, and determining the classification of the symptom based on the comparison. For example, to determine the classification based on the comparison, method 600 may include determining whether the difference between the biological age and the known chronological age exceeds a threshold. If the difference exceeds the threshold, the classification may indicate that the subject has a symptom. The difference can be positive (age acceleration) or negative (age deceleration), each corresponding to a type of age abnormality.

[0178] V. Genes with similar morphological characteristics

[0179] As described in Sections I-III, there may be sets of CpG sites and / or genes whose methylation levels (e.g., 5mC and / or 5hmC levels) change over time. Typical patterns of methylation level changes in these sets of CpG sites and / or genes relative to age (e.g., methylation level trajectories) can be estimated based on methylation level data from subjects without cancer. Methylation level trajectories can then be used to understand biological processes (e.g., disease progression or aging).

[0180] A. Methylation level trajectory during aging

[0181] To estimate the methylation level trajectory during aging (e.g., 5mC level trajectory and / or 5mC level trajectory), z-scoring of cfDNA methylation levels can be performed. For example, z-scoring of the cfDNA 5hmC level for each gene across a first dataset, a second dataset, or a combination thereof can be performed. Alternatively, z-scoring of the methylation level of a set of CpG sites (e.g., the set of CpG sites in Table 4) can be performed. For example, z-scoring of the methylation level of a set of CpG sites obtained by whole-genome bisulfite sequencing of plasma cfDNA samples from 136 cancer-free subjects can be performed.

[0182] Figure 7 The 5hmC level z-score for each highly variable gene identified based on the group is displayed. The 5hmC level z-score for each highly variable gene in each subject is plotted against the subject's age. As a result, Figure 7 A graph 700 shows the cfDNA 5hmC trajectory of highly variable genes during aging, according to some embodiments of this disclosure.

[0183] Additionally, Locally Weighted Scatter Plot Smoothing (LOESS) regression can be fitted to the set of CpG sites and / or each gene to estimate the trajectory of methylation level changes with age for each set of CpG sites or each gene. Alternatively or additionally, methylation level trajectories can be constructed based on various smoothing algorithms, such as, but not limited to, LOESS, binning smoothing, moving average, spline, penalty spline, wavelet, Savitsky-Golay filter, etc. Methylation level trajectories constructed using smoothing algorithms can reveal different patterns (e.g., trajectories) of change in some sets of CpG sites and / or genes with different methylation level trajectories over age. In some examples, when methylation level trajectories are constructed, sets of CpG sites or genes with similar methylation level trajectories can be grouped. Sets of CpG sites or genes with similar 5hmC trajectories can be grouped by unsupervised clustering using fuzzy C-means (FCM). In one implementation, various clustering algorithms can be used to cluster methylation level trajectories, such as, but not limited to, FCM, k-means clustering, hierarchical clustering, principal component analysis (PCA), autoencoders, deep belief networks (DBN), restricted Boltzmann machines (RBM), hierarchical temporal memory (HTM), convolutional neural networks (CNN), support vector machines (SVM), singular value decomposition (SVD), nonnegative matrix factorization (NMF), etc.

[0184] In one example, highly variable genes identified from the data of this cohort can be grouped into clusters based on their corresponding 5hmC trajectories. For this purpose, the clustering method FCM can be used. This method groups genes with similar trajectories into multiple clusters and outputs a membership value for each gene. The membership value of each gene indicates the degree of similarity between the gene and the cluster to which it belongs. A cutoff membership value can be used to select genes with a high correlation to the cluster. For example, the cutoff membership value could be 1. As a result, in this example, 1,926 genes out of 3,000 highly variable genes can be grouped into 8 clusters.

[0185] Figures 8A to 8D Graphs 800a to 800h depict the eight gene clusters identified in the example. These eight gene clusters can each be characterized by a 5hmC trajectory with similar morphological features. Four morphological classifications are shown, but more classifications can be used. Similar morphological features can include similar slopes, similar slope directions (e.g., positive or negative), similar general morphology (e.g., linear, exponential, quadratic, or logarithmic), or combinations thereof. Figures 8A to 8D This can further reveal the 5hmC trajectories of genes in each cluster across an age range (e.g., from 19 to 82 years). For example, it can be... Figures 8A to 8DThe curves 800a to 800h depict the 5hmC trajectory. Each curve 800a to 800h can include the 5hmC level z-score for each gene in the corresponding cluster. The 5hmC level z-score can be plotted against the true age of each object corresponding to each gene. The cluster size can vary relative to multiple related genes.

[0186] like Figure 8A The first and second clusters, as depicted, may include genes with substantially linear 5hmC trajectories, while the other six clusters may include genes with substantially non-linear 5hmC trajectories. For example, as... Figure 8B As shown, the third and fourth clusters can include genes with a substantially logarithmic 5hmC trajectory. Additionally, as... Figure 8C As shown, clusters five and six can include genes with essentially quadratic 5hmC trajectories. Furthermore, as... Figure 8D As shown, clusters seven and eight may include genes with a substantially exponential 5hmC trajectory. The black lines in each of graphs 800a to 800h represent the average value of all genes in each cluster.

[0187] The appropriate morphological classification can be determined in various ways. For example, data points on gene methylation levels in samples collected at different stages can be fitted to a line, and the morphological classification can be linear if the accuracy (e.g., r-value) is within a threshold. Similar operations can be performed on other functional forms.

[0188] B. The relationship between gene clusters and biological phenotypes

[0189] Gene cluster enrichment analysis (GSEA) can be performed to determine whether gene clusters are associated with one or more biological phenotypes. For example, a GSEA analysis may include determining whether the genes in each cluster are associated with enriched biological pathways, enriched tissues, or combinations thereof. The enriched biological pathways, enriched tissues, or combinations thereof for each cluster can be identified using a responseome database (e.g., including biological pathways), the MalaCards database (or other databases that provide genes and their relationships with diseases), other suitable databases, or combinations thereof.

[0190] Based on GSEA, genes in the first cluster can be identified as representing GPCR signaling and neuronal-related pathways. The biological pathways associated with the first cluster can correspond to the tissue types enriched for the first cluster, such as the brain. Additionally, GSEA analysis of cluster 2 can indicate that the corresponding genes are associated with immune and hemostatic pathways and related tissues (e.g., blood and bone marrow). Genes corresponding to the third cluster can involve metabolic (e.g., bio-oxidation) and drug-induced metabolic pathways (ADME) and related tissues, such as the liver and kidneys. Genes associated with the fifth cluster can also be metabolically related and can be particularly enriched in lipid metabolism pathways and endothelial tissues. Genes in the seventh cluster can be skin-related. For example, based on GSEA, genes in the seventh cluster can be identified as being enriched in collagen pathways and dermal tissues. The eighth cluster can include genes associated with ERBB signaling pathways, which can be protein families (e.g., EGFR and HER2) associated with the development of neurodegenerative diseases and some cancers. Furthermore, based on GSEA analysis, genes in each of the eight clusters can be identified as enriched in brain or neuronal tissues, which may indicate a key role of 5hmC in neural development.

[0191] Figures 9A to 9D The graphs 800a to 800h show the gene-specific cfDNA 5hmC trajectory according to embodiments of this disclosure. That is, Figures 9A to 9D The 5hmC trajectories of genes in each cluster identified using the FCM algorithm are shown. Figures 9A to 9D Further, it includes indications of enriched biological pathways and enriched tissues for genes identified in each cluster.

[0192] C. Gene clusters and age indicator markers

[0193] In addition to clusters associated with specific biological pathways and tissues, senescence stages can also exist associated with clusters exhibiting changes in methylation level trajectories. Senescence stages can be indicated by change points or inflection points in the methylation level trajectories. For the set of CpG sites in a cluster and / or many or most of the methylation level trajectories of genes, change points or inflection points can be apparent. Inflection points can be characterized by an increase or decrease in the slope of the methylation level trajectory of the set of CpG sites in a cluster and / or genes. Furthermore, change points are characterized by a change in the direction of the slope of the methylation level trajectory of the set of CpG sites in a cluster and / or genes (e.g., from positive to negative or from negative to positive).

[0194] Figure 10A The graphs 800a to 800b show the 5hmC trajectories of gene-specific cfDNA with age indicator markers 1002a to 1002d. Figure 10BThe graphs showing the 5hmC trajectories of gene-specific cfDNA with age indicator markers 1002e to 1002h (800c to 800d) are displayed. Figure 10C The graphs 800e to 800f show the 5hmC trajectories of gene-specific cfDNA with age indicator markers 1002i to 1002j. Figure 10D A graph showing the 5hmC trajectory of gene-specific cfDNA with age indicator markers 1002k to 1002n is displayed from 800g to 800h. Age indicator markers 1002a to 1002n can be located at change points or inflection points in each cluster.

[0195] For example, as depicted in curve 800a, the variance of the 5hmC trajectory of genes in the first cluster can decrease around age 30. This change in variance of the 5hmC trajectory around age 30 can be labeled using the first age indicator 1002a. The slope of the 5hmC trajectory in the first cluster can also decrease around age 75, labeled using the second age indicator 1002b. This decrease in variance indicates relatively small changes in the 5hmC level of genes in subjects around ages 30 and 75. Similar decreases in 5hmC trajectory variance can also be observed in the second cluster (e.g., depicted in curve 800b), the third cluster (e.g., depicted in curve 800c), the fourth cluster (e.g., depicted in curve 800d), and the eighth cluster (e.g., depicted in curve 800h) in subjects around age 30. Additionally, a decrease in variance around age 75 can be observed in the first, second, and seventh clusters (e.g., depicted in curve 800g). Therefore, based on the changes in the 5hmC trajectories of the first, second, third, fourth, seventh, and eighth clusters, 30 and 70 years of age can be identified as age groups with low genetic dispersion relative to the 5hmC level.

[0196] Furthermore, clusters four, five, six, seven, and eight can exhibit a point of change around age 50. For example, as indicated by the eighth age indicator 1002h and the tenth age indicator 1002j, the 5hmC levels in clusters four and six can increase until around age 50. Then, after age 50, the 5hmC levels of genes in clusters four and six can decrease. The increase or decrease in the 5hmC level of genes can be indicated by the positive or negative slope of the 5hmC trajectory, respectively. Additionally, as indicated by the ninth age indicator 1002i, the eleventh age indicator 1002k, and the fourteenth age indicator 1002n, the 5hmC levels of genes in clusters five, seven, and eight can decrease in subjects under age 50. Then, the 5hmC levels of genes in clusters five, seven, and eight can increase in subjects over age 50.

[0197] Therefore, based on the 5hmC trajectory of clustered genes, the aging stages in which the 5hmC level fluctuates may occur around the ages of 30, 50, and 70.

[0198] Figure 11 Also shown are plots 1100a to 1100c of the 5hmC trajectory of cfDNA during aging according to embodiments of the present disclosure. Figure 11 The cfDNA 5hmc level trajectory depicted in the study represents the total cfDNA 5hmc level in subjects of different ages, rather than the gene-specific 5hmc level. For example... Figure 11 As shown, genome-wide 5hmC levels of cfDNA may show peaks in subjects aged 30, 50, and 70 years. These peaks can be further confirmed using data from two other cohorts (Shao et al., 2022; Walker et al., 2022). Genome-wide 5hmC levels from these two other cohorts are shown in plots 1100b to 1100c.

[0199] D. Organ-specific aging analysis

[0200] Plasma cfDNA analysis can further allow for the assessment of organ aging patterns. In some implementations, cells from the liver, lung, colon, or other suitable organs, in addition to blood cells, can be used for organ-specific aging analysis. To perform organ-specific aging analysis, tissue-specific methylation trajectories (e.g., tissue-specific 5hmC trajectories) can be determined. For example, a group of genes with similar morphological characteristics can be identified, and the relevant tissues can be identified as described herein. The biological age of the subject or the biological age of the corresponding pathway can then be predicted using the tissue-specific methylation trajectories.

[0201] 1. Example model for organ-specific age prediction

[0202] cfDNA methylation level data can be used to estimate biological age associated with specific organs or pathways. For example, as shown in Figure 9, using GSEA assays, clusters one and eight (e.g., corresponding to curves 800a and 800h) can include genes enriched in the brain or neural tissue. Therefore, genes corresponding to clusters one and eight can be considered brain-specific genes. In the example shown in Figure 9, 516 genes corresponding to brain or neural tissue exist between clusters one and eight.

[0203] In the example shown in Figure 9, clusters one and eight are established using unsupervised clustering with fuzzy C-means (FCM). In other embodiments, brain-specific or other organ-specific genes can be defined in a supervised manner. For example, a brain-specific gene can be defined as a gene that displays a differential 5hmC pattern in the brain relative to other tissues. In another example, a brain-specific gene can be defined as a gene that displays differential expression, methylation, or histone modification patterns in the brain relative to other tissues. For example, organ-specific genes with linearly varying methylation levels can be selected.

[0204] Brain-specific aging models can be generated using cfDNA 5hmC data corresponding to genes in clusters one and eight. In one example, the brain-specific aging model can predict age using LASSO regression based on cfDNA 5hmC data. The cfDNA 5hmC data can include the 5hmC level of each gene in clusters one and eight. The cfDNA data can also come from subjects without cancer in the first dataset and can be split into training and test datasets with, for example, a 9:1 ratio.

[0205] In one example, brain-specific genes with detectable 5hmC signals in the second dataset can be used to train and test the brain-specific aging model. Therefore, in this example, 462 genes from clusters one and eight are available as input features for the brain-specific aging model. During training, in this example, the brain-specific aging model can ultimately select a subset of genes (e.g., 180) that can be used to predict the age of an organism's brain.

[0206] After training, the 5hmC levels of genes selected from the test dataset can be input into a brain-specific aging model. The brain-specific aging model can then output a predicted age that can be correlated with the true age of the corresponding cancer-free subject.

[0207] Figure 12A The diagram shows graphs comparing brain-specific age prediction with actual age according to some embodiments of this disclosure. Specifically, Figure 12A This can include the predicted age of a control group (e.g., the actual age of a subject in the test dataset who does not have cancer) compared to the control group's actual age. The predicted age from a brain-specific aging model can be strongly correlated with the actual age. For example, the Pearson coefficient between predicted and actual age could be 0.84.

[0208] Additionally, after training, the 5hmC levels of genes selected from the second dataset can be input into the brain-specific aging model. Therefore, this model can be applied to subjects with gliomas. The brain-specific aging model can then output the predicted age of subjects with gliomas based on the 5hmC levels of genes selected from the second dataset. The predicted age of subjects with gliomas can have a low correlation with the subject's true age. For example, the Pearson coefficient could be 0.40.

[0209] In addition, such as Figure 12A As shown, the predicted age of subjects with gliomas is often greater than their actual age. Therefore, it can be predicted that subjects with gliomas may exhibit accelerated aging relative to controls. A Wilcoxon rank-sum test can be performed to determine whether the accelerated aging in subjects with gliomas is significantly different from that in subjects without cancer.

[0210] Figure 12B Box plots of organ-specific age acceleration according to some embodiments of this disclosure are shown. Figure 12B As shown, the median age acceleration predicted by brain-specific age was 25.07 years for subjects with glioma, while the median age acceleration predicted by brain-specific age was -0.99 years for subjects without cancer. Furthermore, according to the Wilcoxon rank-sum test, p-values ​​below 0.001 indicate a significant difference in age acceleration between the two groups.

[0211] and Figures 5A to 5B In contrast, the median age acceleration associated with brain-specific age prediction can be greater than the median age acceleration associated with age predictions without organ-specific predictions. This may suggest that brain-specific aging clocks (e.g., brain-specific aging models) can contain enhanced information related to the effects of gliomas on the brain or neural tissue.

[0212] 2. Organ-specific aging clock

[0213] Comparing the biological age and chronological age of a specific organ can allow for the determination of an acceleration or deceleration of the organ's biological age, potentially enabling the detection and preventative intervention of non-invasive diseases associated with that organ. To further demonstrate the feasibility of establishing an aging clock for individual organs, organ-specific aging models for K different organs were constructed using organ-specific genes exhibiting elevated 5hmC levels. As an example, K can be at least 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, or more organs. For example, a brain-specific gene can refer to a gene with a high methylation level (e.g., high 5hmC level) in brain tissue relative to other tissue types. In another example, an organ or tissue-specific gene can be defined as a gene that exhibits a differential pattern of 5mC, methylation, expression, or histone modification relative to other organs or tissues.

[0214] In this example, K (e.g., 15) organ-specific aging models were developed using LASSO regression based on the cfDNA 5hmC data of subjects without cancer in the first dataset. Similar to the brain-specific aging model, each organ-specific model obtained 5hmC data for genes enriched in biological pathways, tissues, or other organ-related genes. The cfDNA 5hmC data corresponding to each gene for each organ could be split into training and testing datasets. Figures 13A to 13D The model can depict the predicted age provided by each organ-specific model using the test and training datasets for each organ. Figures 13A to 13D The predicted age can be plotted against the true age of cancer-free subjects from the first dataset.

[0215] The Pearson coefficient can be used to estimate the accuracy of the predicted age of each organ-specific model relative to the true age of subjects without cancer. Figure 14 Pearson correlation coefficients for each organ-specific age model according to some embodiments of this disclosure are shown in plot 1400. These organ-specific models can provide in-depth understanding of the aging process at the organ level, which can allow for more precise assessment of the aging process or disease progression.

[0216] In some implementations, the age of two or more organs can be determined. These organ ages can be combined to determine an organ-specific age spectrum or pattern based on different age-specific models (e.g., liver age, brain age, etc.). The total age derived from various genes can also be determined and used in an age spectrum (e.g., a set). This person's age spectrum can be compared to that of a healthy person or someone with known disease.

[0217] E. Methods using morphological characteristics

[0218] Figure 15 This is a flowchart illustrating a method 1500 for detecting lesions (e.g., symptoms or conditions) in an object of known actual age, according to some embodiments of this disclosure. Some aspects of method 1500 can be performed in a manner similar to methods 400 and 600.

[0219] At box 1502, method 1500 may include receiving sequence reads comprising the methylation state at sites on a plurality of cell-free DNA (cfDNA) molecules. The sequence reads may include one or more sites on each cfDNA molecule. Box 1502 may be performed in a manner similar to box 402 of method 400.

[0220] At box 1504, method 1500 may include aligning a sequence reading with a reference genome. The reference genome may be a representative example of a DNA sequence corresponding to a species of the object. The alignment may involve searching for the best match for the sequence reading in the reference genome.

[0221] Sequence reads can be aligned to a set of CpG sites in the genome. Table 4 provides examples of CpG sites that can be aligned to sequence reads. Alternatively, sequence reads can be aligned to genes in a reference genome. Genes may include those corresponding to… Figure 2 Any number of the 3,000 genes depicted in the diagram and provided in Table 1. In some examples, at least some of the sets of one or more CpG sites may correspond to a gene. Therefore, at least some sequence reads can be aligned to the set of CpG sites within a gene. Table 4 shows examples of CpG sites corresponding to genes. Box 1504 can be performed in a manner similar to box 404 of method 400.

[0222] At box 1506, method 1500 may include, for each of one or more sets of CpG sites, identifying a set of sequence readings aligned to any CpG site in that set of CpG sites. The set of CpG sites may include at least three sets of CpG sites. In some examples, the set of CpG sites may include at least 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, or 250 CpG sites. In addition, in some examples, a set of CpG sites may include a set of at least 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 30, 40, 50, 60, 70, 80, 90, or 100 CpG sites.

[0223] Furthermore, each set of CpG sites within this group can exhibit the same morphological classification for changes in methylation levels over age. This morphological classification can be linear, logarithmic, quadratic, or exponential. Therefore, each CpG site within each set of CpG sites in this group can exhibit similar changes in methylation levels over time in a substantially linear, logarithmic, quadratic, or exponential manner.

[0224] Alternatively, as mentioned above, one or more sets of CpG sites may correspond to genes. In such examples, a set of CpG sites may correspond to a set of genes, where each gene in each set may have the same morphological classification regarding changes in methylation levels over time. The morphological classification may be linear, logarithmic, quadratic, or exponential. That is, each gene in the set may exhibit similar changes in methylation levels over time in a substantially linear, logarithmic, quadratic, or exponential manner. For example, this set of genes may correspond to the clusters discussed above with respect to Figures 8 through 10.

[0225] In some examples, each set of CpG sites and / or each gene corresponding to each set of CpG sites can be associated with a specific tissue. The association of each set of CpG sites or each gene with a specific tissue can be determined based on (1) a biological pathway or (2) an epigenetic pattern or (3) an expression level above a threshold in a specific tissue type. A specific tissue type can be an organ. More specifically, tissue types can include bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, or testis.

[0226] At box 1508, method 1500 may include determining one or more methylation levels for each group in a set of CpG sites using the methylation status of the sequence reads from that group. In some examples, the methylation level may be the number of methylated sites in each set of CpG sites compared to a set of sequence reads. Alternatively, in some examples, the methylation level may be a proportion, percentage, or other suitable concentration quantification of methylated sites in each set of CpG sites. That is, the methylation level may be a measure of the number of methylated sites in each set of CpG sites relative to the total number of sites in the group. As an example, the number of groups of CpG sites (e.g., genes) may be at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 30, 40, or 50 groups. Each gene may include a variety of numbers of CpG sites, such as the numbers listed herein for other methods.

[0227] For example, to determine the methylation level (e.g., 5hmC level and / or 5mC level) of a set of CpG sites, the amount (number) of methylation states indicating the presence or absence of methylation at that set of one or more CpG sites can be determined. Therefore, in some examples, the methylation level of a set of sequence readings aligned to a set of CpG sites could be the amount of methylation states indicating the presence of methylation at each CpG site in that set.

[0228] Alternatively, in at least some of the examples corresponding to genes within the set of one or more CpG sites, the methylation level can be a gene-specific methylation level. For example, the methylation level can be the amount of methylation state indicating the presence of methylation corresponding to each CpG site of a gene.

[0229] At box 1510, method 1500 may include using an age-dependent model to determine the classification of the presence of lesions in an object. The classification of the presence of lesions may be determined using the object's known actual age and one or more methylation levels. The model may be an age-dependent machine learning model. The age-dependent machine learning model may include multiple sub-models, each corresponding to a different actual age. The model may be generated using reference samples of objects with known lesion classifications.

[0230] In one example, determining the classification may include comparing one or more methylation levels to one or more thresholds, wherein the one or more thresholds depend on the known actual age. For example, the one or more thresholds may use a normal range for a given actual age (e.g., within a specified number of standard deviations).

[0231] In some examples, determining the classification of lesions present in an object may include generating a feature vector based on one or more methylation levels of the set of CpG sites, and comparing the feature vector with a representative reference vector determined using a set of reference vectors measured to have the same known lesion classification (e.g., disease classification). The distance between the two vectors can be determined. If the feature vector is sufficiently similar to a first representative reference vector (e.g., the distance is within a threshold or closest to the representative reference vector), the object can be determined to have a first classification corresponding to the first representative reference vector.

[0232] The model can generate a first representative reference vector based on reference vectors of sample subjects who have lesions (e.g., gliomas) and are the same age as or similar to the subjects (e.g., within the same age range). Therefore, the first representative reference vector can include a set of CpG sites and / or measured (including statistically determined, e.g., averaged) methylation levels of genes for the sample subjects with lesions.

[0233] Additionally, the model can generate a second reference vector for healthy samples of subjects of the same or similar age. Therefore, predicting the presence of lesions in a subject could involve comparing the feature vector to each representative reference vector and determining which representative reference vector the feature vector is most similar to. In some examples, there could be any number of representative reference vectors corresponding to any number of lesions.

[0234] Each actual age or age range can have its own set of representative feature vectors. Therefore, a model can be considered as comprising sub-models for different ages, where each sub-model has a set of representative feature vectors for its respective classification. In other implementations, the model can determine the distances to all ages relative to all other representative reference vectors at once; however, the distances to these reference vectors for different ages can be weighted by scaling that dimension, such that the distances to the reference vectors for different ages are generally higher than the distances to the reference vectors for the same age.

[0235] Additionally, in some examples, determining the classification of a lesion's presence may include using a model to predict the subject's biological age, comparing the biological age to a known chronological age, and determining the classification of the lesion's presence based on the comparison. For example, to determine the classification based on the comparison, method 1500 may include determining whether the difference between the biological age and the known chronological age exceeds a threshold. The threshold may be based on the subject's known chronological age. If the difference exceeds the threshold, the classification may indicate that the subject has a lesion (e.g., glioma). The difference may be positive (accelerated aging) or negative (decelerated aging), each corresponding to a type of age abnormality.

[0236] Determining the classification of lesions may also include using an age-dependent machine learning model to predict the subject's biological age. Method 1500 may then include comparing the biological age with the subject's known actual age and determining, based on the comparison, the classification of lesions present in the subject, wherein the lesion is a lesion specific to a particular tissue type.

[0237] VI. Age accuracy for different numbers of features

[0238] The accuracy of age predictions provided by the model can be determined for different numbers of features. In one example, the number of different features could be the number of genes. In another example, the number of different features could be the number of CpG sites.

[0239] Figure 17A Pearson correlation coefficients between the actual and predicted ages in the training and test datasets are shown in graph 1700. These training and test datasets were used to obtain cfDNA 5hmC data and, as described above, for... Figure 3 The 785 objects described in the text are associated. Pearson correlation coefficients are plotted against the number of features used in the corresponding age prediction model (e.g., the linear regression model described above) to predict age. Additionally, Figure 17B Table 1702 shows the Pearson correlation coefficient and the number of features from graph 1700.

[0240] In Plot 1700, the feature number is the number of genes used by the age prediction model to predict age. The genes used in the age prediction models correlated with the Pearson correlation coefficient in Plot 1700 were selected from 3,000 genes with the highest normalized standard deviation values, such as... Figure 2 The solid black dots in the image are shown. In other examples, any number of genes (e.g., at least 20, 30, 40, 50, 70, 80, 90, 100, 150, 250, etc.) can be used in age prediction models to predict age based on the gene's cytosine methylation level.

[0241] Furthermore, in plot 1700, line 1704 shows the trajectory of the Pearson correlation coefficient with respect to the number of genes for age prediction on the training dataset. In contrast, line 1706 shows the trajectory of the Pearson correlation coefficient with respect to the number of genes for age prediction on the test dataset. In one example, as shown in plot 1700 and table 1702, when the cytosine methylation levels of 100 genes are input into the age prediction model, the Pearson correlation coefficients for the training and test datasets are 0.9 and 0.91, respectively.

[0242] Figure 18AAlso shown is a graph 1800 showing the Pearson correlation coefficients between the actual and predicted ages in the training and test datasets associated with 136 objects. The Pearson correlation coefficients are plotted against the number of features used in the respective age prediction models (e.g., the linear regression model described above). Figure 18B Table 1802 shows the Pearson correlation coefficient and number of features from graph 1800.

[0243] Compared to plot 1700, the number of features in plot 1800 is determined by the number of CpGs used in the corresponding age prediction model. Therefore, line 1804 shows the trajectory of the Pearson correlation coefficient for age prediction on the training dataset relative to the corresponding number of CpGs. Then, line 1806 shows the trajectory of the Pearson correlation coefficient for predicted age on the test dataset relative to the corresponding number of CpGs. In one example, as shown in plot 1800 and table 1802, when the cytosine methylation levels of 30 CpGs are input into the age prediction model, the Pearson correlation coefficients for the training and test datasets are 0.96 and 0.94, respectively.

[0244] The CpGs used in the age prediction model associated with curve 1800 were selected from the 49 CpGs mentioned above regarding the aging clock based on cytosine methylation levels. Any number of CpGs (e.g., at least 5, 10, 20, 30, 40, 49, 50, etc.) can be used in age prediction models to predict age based on CpG cytosine methylation levels.

[0245] VII. Example System

[0246] Figure 20 A measurement system 2000 according to one embodiment of the present disclosure is shown. The system includes a sample 2005, such as a free DNA molecule (e.g., DNA and / or RNA), within an analytical apparatus 2010, through which an analysis 2008 can be performed on the sample 2005. For example, the sample 2005 can be contacted with reagents of the analysis 2008 to provide a signal (e.g., intensity signal) of a physical characteristic 2015 (e.g., sequence information of a free nucleic acid molecule). An example of the analytical apparatus may be a flow channel comprising a probe and / or primer or droplet through which the analysis moves (where the droplet includes the analysis). A detector 2020 detects a physical characteristic 2016 of the sample (e.g., fluorescence intensity, voltage, or current). The detector 2020 can perform measurements at time intervals (e.g., periodic time intervals) to obtain data points constituting a data signal. In one embodiment, an analog-to-digital converter converts the analog signal from the detector into digital form at multiple times.

[0247] The analysis device 2010 and detector 2020 can form an analysis system, such as a sequencing system that performs sequencing according to the embodiments described herein. Data signal 2025 is transmitted from detector 2020 to logic system 2030. As an example, data signal 2025 can be used to determine the sequence and / or location of a nucleic acid molecule (e.g., DNA and / or RNA) in a reference genome. Data signal 2025 can include various measurements taken simultaneously, such as different colors of fluorescent dyes or different electrical signals for different molecules in sample 2005, and therefore data signal 2025 can correspond to multiple signals. Data signal 2025 can be stored in local memory 2035, external memory 2040, or storage device 2045. The analysis system can consist of multiple analysis devices and detectors.

[0248] The logic system 2030 may be a computer system, ASIC, microprocessor, graphics processing unit (GPU), etc., or may include a computer system, ASIC, microprocessor, graphics processing unit (GPU), etc. It may also include a display (e.g., a monitor, LED display, etc.) and user input devices (e.g., a mouse, keyboard, buttons, etc.), or be connected to a display and user input devices. The logic system 2030 and other components may be part of a stand-alone or network-connected computer system, or may be directly connected to or incorporated into a device including detector 2020 and / or analysis device 2010 (e.g., a sequencing device). The logic system 2030 may also include software executing in processor 2050. The logic system 2030 may include a computer-readable medium storing instructions for controlling measurement system 2000 to perform any of the methods described herein. For example, the logic system 2030 may provide commands to a system including analysis device 2010 to perform sequencing or other physical operations. Such physical operations may be performed in a specific order, such as adding and removing reagents in a specific order. Such physical operations can be performed using robotic systems (such as robotic arms) that can be used to obtain samples and perform analysis.

[0249] The measurement system 2000 may also include a treatment device 2060 capable of providing treatment to an object. The treatment device 2060 may determine a treatment and / or be used to perform a treatment. Examples of such treatments may include surgery, radiation therapy, chemotherapy, immunotherapy, targeted therapy, hormone therapy, and stem cell transplantation. A logic system 2030 may be connected to the treatment device 2060, for example, to provide results of the methods described herein. The treatment device may receive input from, for example, imaging devices and other devices (e.g., for controlling the treatment, such as controlling a robotic system).

[0250] Any computer system mentioned in this article can utilize any suitable number of subsystems. Examples of such subsystems are shown in [the document / examples]. Figure 21 In computer system 10. In some embodiments, the computer system includes a single computer device, wherein a subsystem may be a component of the computer device. In other embodiments, the computer system may include multiple computer devices having internal components, each of which is a subsystem. The computer system may include desktop and laptop computers, tablet computers, mobile phones, and other mobile devices.

[0251] Shown Figure 21 The subsystems in the display are interconnected via system bus 75. Additional subsystems, such as printer 74, keyboard 78, one or more storage devices 79, monitor 76 (e.g., display screen, such as LED) connected to display adapter 82, and other devices, are displayed. Peripheral devices and I / O devices connected to input / output (I / O) controller 71 can be connected to the computer system via a variety of devices known in the art, such as input / output (I / O) port 77 (e.g., USB, FireWire®). For example, I / O port 77 or external interface 81 (e.g., Ethernet, Wi-Fi, etc.) can be used to connect computer system 10 to a wide area network (e.g., the Internet), a mouse input device, or a scanner. The interconnection via system bus 75 allows central processing unit 73 to exchange information with each subsystem and control the execution of multiple instructions from system memory 72 or storage device 79 (e.g., fixed disk, such as hard disk drive, or optical disk), as well as information exchange between subsystems. System memory 72 and / or storage device 79 may include computer-readable media. Another subsystem is the data collection device 85, such as a camera, microphone, accelerometer, etc. Any data mentioned herein can be output from one component to another and can also be output to the user.

[0252] A computer system may include multiple identical components or subsystems, connected together, for example, via an external interface 81, an internal interface, or via a removable storage device that can be connected to and moved from one component to another. In some embodiments, the computer system, subsystem, or device may communicate via a network. In such cases, one computer may be considered a client and another a server, where each computer may be part of the same computer system. Clients and servers may each include multiple systems, subsystems, or components. In various embodiments, the method may involve a variety of numbers of clients and / or servers, including at least 10, 20, 50, 100, 200, 500, 1,000, or 10,000 devices. The method may include a variety of numbers of communication messages between devices, including at least 100, 200, 500, 1,000, 10,000, 50,000, 100,000, 500,000, or one million communication messages. Such communications can involve at least 1 MB, 10 MB, 100 MB, 1 GB, 10 GB, or 100 GB of data.

[0253] Aspects of the implementation may be implemented using hardware circuitry (e.g., application-specific integrated circuits or field-programmable gate arrays) and / or using computer software stored in a modular or integrated manner in memory with a general-purpose programmable processor in the form of logical control, and thus the processor may include memory storing software instructions for configuring the hardware circuitry and an FPGA or ASIC having configuration instructions. As used herein, the processor may include a single-core processor, a multi-core processor on the same integrated chip, or single board or network hardware, and multiple processing units on dedicated hardware. Based on this disclosure and the teachings provided herein, those skilled in the art will appreciate other ways and / or methods of implementing embodiments of this disclosure using hardware and combinations of hardware and software.

[0254] Any software component or function described in this application may be implemented in the form of software code using, for example, conventional or object-oriented techniques. This software code is executed by a processor using any suitable computer language (e.g., Java, C, C++, C#, Objective-C, Swift) or scripting language (e.g., Perl or Python). The software code may be stored as a series of instructions or commands on a computer-readable medium for storage and / or transmission. Suitable non-transitory computer-readable media may include random access memory (RAM), read-only memory (ROM), magnetic media (e.g., hard disk or floppy disk) or optical media (e.g., closed-disk CD, digital optical disc (DVD), or Blu-ray disc), flash memory, and similar devices. The computer-readable medium may be any combination of such devices. Additionally, the order of operations may be reconfigured. A process may terminate upon completion of its operations, but may have additional steps not included in the figure. A process may correspond to a method, function, program, subroutine, subroutines, etc. When a process corresponds to a function, its termination may correspond to the function returning to the calling function or the main function.

[0255] Such programs can also be encoded and transmitted using carrier signals suitable for transmission over wired, optical, and / or wireless networks (including the Internet) conforming to various protocols. Therefore, computer-readable media can be constructed using data signals encoded by such programs. Computer-readable media encoded with program code (e.g., as software) can be packaged with compatible devices or provided separately from other devices (e.g., for download via the Internet). Any such computer-readable media can reside on or within a single computer product (e.g., a hard disk drive, CD, or an entire computer system) and can reside on or within different computer products within a system or network. A computer system may include a monitor, printer, or other suitable display for providing a user with any of the results mentioned herein.

[0256] Any method described herein can be performed wholly or partially using a computer system including one or more processors configured to perform the steps described. Any operation performed via the processor (e.g., comparison, determination, contrast, computer calculation, computation) can be performed in real time. The term "..." real time"This can refer to a computational operation or process completed within a certain time limit. The time limit can be 1 minute, 1 hour, 1 day, or 7 days. Therefore, an implementation can refer to a computer system configured to perform the steps of any method described herein, potentially having different components that perform the respective steps or groups of steps. Although presented in the form of numbered steps, the steps of the methods herein can be performed simultaneously or at different times or in different orders. Furthermore, a portion of these steps can be used in conjunction with a portion of other steps of other methods. Additionally, all or some steps can be optional. Moreover, any step of any method can be performed using modules, units, circuits, or other means of a system for performing these steps."

[0257] The specified details of a particular embodiment may be combined in any suitable manner without departing from the spirit and scope of the embodiments of this disclosure. However, other embodiments of this disclosure may refer to specific embodiments relating to each individual aspect or a particular combination of these individual aspects.

[0258] The foregoing description of exemplary embodiments of this disclosure has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit this disclosure to the precise form described, and many modifications and variations are possible in light of the foregoing teachings.

[0259] Unless explicitly stated otherwise, the use of “a,” “an,” or “the” is intended to mean “one or more.” Unless explicitly stated otherwise, the use of “or” is intended to mean “inclusive of or” rather than “exclusive of or.” Referring to the “first” component does not necessarily require the provision of the second component. Furthermore, unless explicitly stated otherwise, referring to the “first” or “second” component does not limit the referenced component to a particular location. The term “based on” is intended to mean “at least partially based on.”

[0260] The claims are drafted to exclude any optional elements. Therefore, this statement is intended to serve as a basis for the use of exclusive terms such as “solely,” “only,” and similar terms, or for the use of “negative” limitations, in conjunction with the description of the claimed elements.

[0261] All patents, patent applications, publications, and specifications mentioned herein are incorporated herein by reference in their entirety for all purposes. No prior art is acknowledged. In the event of any conflict between this application and the references provided herein, this application shall prevail.

[0262] VIII. Tables

[0263] The genes for the eight clusters (groups) in the above embodiments are provided in Table 1 below. Various implementations may use all or fewer of the genes for any given cluster.

[0264]

[0265] Table 1: Corresponding to those from Figures 8A to 8D The genes of the 8 clusters.

[0266] The genes used in organ-specific aging models are provided in Table 2 below. Various implementation schemes may use all or fewer of the genes from any given cluster.

[0267]

[0268] Table 2: Corresponding to Figures 13A to 13D The genes of the organs in the body.

[0269] The genes used in the aging model are provided in Table 3 below. Various implementation schemes may use all or fewer of the genes.

[0270]

[0271] Table 3: Corresponding to Figures 16 to 1 The genes used in data 9.

[0272] The CpG sites used in the aging model are provided in Table 3 below. Various implementation schemes may use all or fewer of the CpG sites.

[0273]

[0274] Table 4: Corresponding to Figures 16 to 1 The CpG sites and corresponding genomic locations used in 9.

[0275] IX. References

[0276] Cai, J., Chen, L., Zhang, Z., Zhang, X., Lu, X., Liu, W., Shi, G., Ge, Y., Gao, P., and Yang, Y. (2019). Genome-wide mapping of 5-hydroxymethylcytosines in circulating cell-free DNA as a non-invasive approach for early detection of hepatocellular carcinoma. Gut 68, 2195-2205.

[0277] Cai, J., Zeng, C., Hua, W., Qi, Z., Song, Y., Lu, X., Li, D., Zhang, Z., Cui, X., and Zhang, X. (2021). An integrative analysis of genome-wide 5-hydroxymethylcytosines in circulating cell-free DNA detects noninvasive diagnostic markers for gliomas. Neuro-oncology advances 3, vdab049.

[0278] Fleischer, J.G., Schulte, R., Tsai, H.H., Tyagi, S., Ibarra, A., Shokhirev, M.N., Huang, L., Hetzer, M.W., and Navlakha, S. (2018). Predicting age from the transcriptome of human dermal fibroblasts. Genome biology 19, 1-8.

[0279] Greenberg, M.V., and Bourc’his, D.(2019). The diverse roles of DNAmethylation in mammalian development and disease. Nature reviews Molecularcell biology 20, 590-607.

[0280] Hannum, G., Guinney, J., Zhao, L., Zhang, L., Hughes, G., Sadda, S.,Klotzle, B., Bibikova, M., Fan, J.-B., and Gao, Y.(2013). Genome-widemethylation profiles reveal quantitative views of human aging rates.Molecular cell 49, 359-367.

[0281] Hertel, J., Friedrich, N., Wittfeld, K., Pietzner, M., Budde, K., Vander Auwera, S., Lohmann, T., Teumer, A., V lzke, H., and Nauck, M.(2016).Measuring biological age via metabonomics: the metabolic age score. Journalof proteome research 15, 400-410.

[0282] Horvath, S.(2013). DNA methylation age of human tissues and celltypes. Genome biology 14, 1-20.

[0283] Koh, K.P., Yabuuchi, A., Rao, S., Huang, Y., Cunniff, K., Nardone,J., Laiho, A., Tahiliani, M., Sommer, C.A., and Mostoslavsky, G.(2011). Tet1and Tet2 regulate 5-hydroxymethylcytosine production and cell lineagespecification in mouse embryonic stem cells. Cell stem cell 8, 200-213.

[0284] Kriaucionis, S., and Heintz, N.(2009). The nuclear DNA base 5-hydroxymethylcytosine is present in Purkinje neurons and the brain. Science324, 929-930.

[0285] Lehallier, B., Gate, D., Schaum, N., Nanasi, T., Lee, S.E., Yousef,H., Moran Losada, P., Berdnik, D., Keller, A., and Verghese, J.(2019).Undulating changes in human plasma proteome profiles across the lifespan.Nature medicine 25, 1843-1850.

[0286] Leon, S., Shapiro, B., Sklaroff, D., and Yaros, M.(1977). Free DNA inthe serum of cancer patients and the effect of therapy. Cancer research 37,646-650.

[0287] Li, J., Xiong, M., Fu, X.-H., Fan, Y., Dong, C., Sun, X., Zheng, F.,Wang, S.-W., Liu, L., and Xu, M.(2023). Determining a multimodal aging clockin a cohort of Chinese women. Med 4, 825-848. e813.

[0288] Liao, P., Ostrom, Q.T., Stetson, L., and Barnholtz-Sloan, J.S.(2018).Models of epigenetic age capture patterns of DNA methylation in gliomaassociated with molecular subtype, survival, and recurrence. Neuro-oncology20, 942-953.

[0289] Lin, Q., and Wagner, W.(2015). Epigenetic aging signatures arecoherently modified in cancer. PLoS genetics 11, e1005334.

[0290] Lister, R., Mukamel, E.A., Nery, J.R., Urich, M., Puddifoot, C.A.,Johnson, N.D., Lucero, J., Huang, Y., Dwork, A.J., and Schultz, M.D.(2013).Global epigenomic reconfiguration during mammalian brain development. Science341, 1237905.

[0291] Lo, Y.D., Corbetta, N., Chamberlain, P.F., Rai, V., Sargent, I.L.,Redman, C.W., and Wainscoat, J.S.(1997). Presence of fetal DNA in maternalplasma and serum. The lancet 350, 485-487.

[0292] Lo, Y.D., Tein, M.S., Pang, C.C., Yeung, C.K., Tong, K.-L., andHjelm, N.M.(1998). Presence of donor-specific DNA in plasma of kidney andliver-transplant recipients. The Lancet 351, 1329-1330.

[0293] Lowsky, D.J., Olshansky, S.J., Bhattacharya, J., and Goldman, D.P.(2014). Heterogeneity in healthy aging. Journals of Gerontology Series A:Biomedical Sciences and Medical Sciences 69, 640-649.

[0294] Mandel, P.(1948). Les acides nucleiques du plasma sanguin chez 1homme. CR Seances Soc Biol Fil 142, 241-243.

[0295] Occean, J.R., Yang, N., Sun, Y., Dawkins, M.S., Munk, R., Belair, C.,Dar, S., Anerillas, C., Wang, L., and Shi, C.(2023). Gene body DNAhydroxymethylation restricts the magnitude of transcriptional changes duringaging. bioRxiv.

[0296] Oh, H.S.-H., Rutledge, J., Nachun, D., Pálovics, R., Abiose, O.,Moran-Losada, P., Channappa, D., Urey, D.Y., Kim, K., and Sung, Y.J.(2023).Organ aging signatures in the plasma proteome track health and disease.Nature 624, 164-172.

[0297] Partridge, L., Deelen, J., and Slagboom, P.E.(2018). Facing up to theglobal challenges of ageing. Nature 561, 45-56.

[0298] Pastor, W.A., Pape, U.J., Huang, Y., Henderson, H.R., Lister, R., Ko,M., McLoughlin, E.M., Brudno, Y., Mahapatra, S., and Kapranov, P.(2011).Genome-wide mapping of 5-hydroxymethylcytosine in embryonic stem cells.Nature 473, 394-397.

[0299] Peters, M.J., Joehanes, R., Pilling, L.C., Schurmann, C., Conneely,K.N., Powell, J., Reinmaa, E., Sutphin, G.L., Zhernakova, A., and Schramm, K.(2015). The transcriptional landscape of age in human peripheral blood.Nature communications 6, 1-14.

[0300] Robinson, O., Chadeau Hyam, M., Karaman, I., Climaco Pinto, R., Ala-Korpela, M., Handakas, E., Fiorito, G., Gao, H., Heard, A., and Jarvelin,M.R.(2020). Determinants of accelerated metabolomic and epigenetic aging in aUK cohort. Aging Cell 19, e13149.

[0301] Rutledge, J., Oh, H., and Wyss-Coray, T.(2022). Measuring biologicalage using omics data. Nature Reviews Genetics 23, 715-727.

[0302] Satija, R., Farrell, J.A., Gennert, D., Schier, A.F., and Regev, A.(2015). Spatial reconstruction of single-cell gene expression data. Naturebiotechnology 33, 495-502.

[0303] Shao, J., Bernicker, E.H., He, C., and Li, Z.(2022). Cell‐free DNA 5‐hydroxymethylcytosine as a marker for common cancer detection. Clinical andTranslational Discovery 2, e136.

[0304] Szulwach, K.E., Li, X., Li, Y., Song, C.-X., Wu, H., Dai, Q., Irier,H., Upadhyay, A.K., Gearing, M., and Levey, A.I.(2011). 5-hmC–mediatedepigenetic dynamics during postnatal neurodevelopment and aging. Natureneuroscience 14, 1607-1616.

[0305] Walker, N.J., Rashid, M., Yu, S., Bignell, H., Lumby, C.K., Livi,C.M., Howell, K., Morley, D.J., Morganella, S., and Barrell, D.(2022).Hydroxymethylation profile of cell-free DNA is a biomarker for earlycolorectal cancer. Scientific Reports 12, 16566.

[0306] Xiong, J., Jiang, H.-P., Peng, C.-Y., Deng, Q.-Y., Lan, M.-D., Zeng,H., Zheng, F., Feng, Y.-Q., and Yuan, B.-F.(2015). DNA hydroxymethylation ageof human blood determined by capillary hydrophilic-interaction liquidchromatography / mass spectrometry. Clinical epigenetics 7, 1-11.

Claims

1. A method for measuring the biological age of an object, the method comprising performing the following via a computer system: Receive sequence reads, which include the methylation status at sites on multiple free DNA molecules; The sequence reads are aligned to a reference genome, wherein the sequence reads are aligned to N sets of one or more CpG sites; For each of the N sets of the one or more CpG sites: Identify a set of sequence reads aligned to one or more CpG sites in the reference genome; and The methylation level is determined using the methylation status of the set of sequence reads, where N is at least 3, thereby determining N methylation levels; Generate a feature vector based on the N methylation levels; The machine learning model is loaded into the memory of the computer system, and the machine learning model is trained using training samples with reference vectors having known actual age and measured methylation levels; The feature vector is input into the machine learning model; and The machine learning model is used to predict the biological age of the object.

2. The method of claim 1, wherein the N sets of one or more CpG sites are associated with a specific tissue type, and wherein the biological age is a biological age specific to the specific tissue type.

3. The method of claim 2, wherein the N sets of one or more CpG sites are associated with the specific tissue type based on (1) a biological pathway, (2) an epigenetic pattern, or (3) an expression level greater than a threshold in the specific tissue type.

4. The method of claim 2, wherein the specific tissue type is a tissue type specific to a specific organ.

5. The method of claim 2, wherein the specific tissue type is selected from Table 2, and wherein the N sets of one or more CpG sites are selected from the genes listed in Table 2 that are associated with the specific tissue type.

6. The method of claim 2, wherein the specific tissue type is selected from: bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis.

7. The method according to claim 1, wherein the biological age is an age range.

8. The method of claim 1, wherein the machine learning model is a regression model.

9. The method according to any one of claims 1 to 8, wherein determining the methylation level at the set of one or more CpG sites comprises determining an amount of methylation state at the one or more CpG sites indicating the presence or absence of methylation.

10. The method of claim 9, wherein the methylation level is methylation density.

11. The method of claim 9, wherein the methylation level comprises a proportion of methylation states at the site indicating the presence or absence of the methylation.

12. The method according to any one of the preceding claims, wherein the N sets of one or more CpG sites correspond to N genes, and wherein the N methylation levels are N gene-specific methylation levels.

13. The method according to claim 12, wherein the N gene-specific methylation level is 5 hmC.

14. A method for detecting lesions in a subject of known actual age, the method comprising execution via a computer system: Receive sequence reads, which include the methylation status at sites on multiple free DNA molecules; The sequence reads are aligned to a reference genome, wherein the sequence reads are aligned to N sets of one or more CpG sites; For each of the N sets of the one or more CpG sites: Identify a set of sequence reads aligned to one or more CpG sites in the reference genome; and The methylation level is determined using the methylation status of the set of sequence reads, where N is at least 3, thereby determining N methylation levels; Generate a feature vector based on the N methylation levels; An age-dependent machine learning model is loaded into the memory of the computer system. The age-dependent machine learning model is trained using training samples with reference vectors having known actual age, known lesion classification, and measured methylation levels. The feature vector is input into the age-dependent machine learning model; and The classification of the presence of lesions in the object is determined by using the age-dependent machine learning model of the feature vectors.

15. The method of claim 14, wherein the known actual age is an age range.

16. The method of claim 14, wherein the age-dependent machine learning model comprises a plurality of sub-models, each corresponding to a different actual age.

17. The method of claim 14, wherein determining the classification of the presence of the lesion in the object comprises: The feature vector is compared with a representative reference vector, which is determined using a set of reference vectors of measurements with the same known lesion classification.

18. The method of claim 14, wherein determining the classification of the presence of the lesion in the object comprises: The age-dependent machine learning model is used to predict the biological age of the object; Compare the biological age with the known actual age; and The classification of the lesions in the object is determined based on the comparison.

19. The method of claim 18, wherein determining the classification of the presence of the lesion in the object further comprises: Determine the difference between the biological age and the known actual age; and The difference is compared with a threshold to determine the classification of the presence of the lesion in the object.

20. The method of claim 14, wherein the N sets of one or more CpG sites are associated with a specific tissue type, and wherein the lesion is a lesion specific to the specific tissue type.

21. The method of claim 20, wherein the N sets of one or more CpG sites are associated with the particular tissue type based on (1) a biological pathway, or (2) an epigenetic pattern, or (3) an expression level greater than a threshold in the particular tissue type.

22. The method of claim 20, wherein the specific tissue type is a tissue type specific to a specific organ.

23. The method of claim 20, wherein the specific tissue type is selected from Table 2, and wherein the N sets of CpG sites are selected from the genes listed in Table 2 that are associated with the specific tissue type.

24. The method of claim 20, wherein the specific tissue type is selected from: bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis.

25. The method of any one of claims 14 to 24, wherein determining the methylation level at the set of one or more CpG sites comprises determining an amount of methylation at the one or more CpG sites that indicates the presence or absence of methylation.

26. The method of claim 25, wherein the methylation level is the methylation density.

27. The method of claim 25, wherein the methylation level comprises the proportion of methylation states at the site that indicate the presence or absence of the methylation.

28. The method according to any one of claims 14 to 27, wherein the N sets of the one or more CpG sites correspond to N genes, and wherein the N methylation levels are N gene-specific methylation levels.

29. The method of claim 28, wherein the N gene-specific methylation levels are at the 5 hmC level.

30. A method for detecting lesions in a subject of known actual age, the method comprising being performed via a computer system: Receive sequence reads, which include the methylation status at sites on multiple free DNA molecules; The sequence readings were compared with a reference genome. For each group in the set of one or more groups of CpG sites: Identify a set of sequence reads that align to any CpG site in the set of CpG sites of the group, the set of CpG sites of the group comprising a set of at least 3 CpG sites, wherein each set of CpG sites in the group has the same morphological classification for changes in methylation levels with age. and The methylation status of the set of sequence readings is used to determine one or more methylation levels; and The classification of the presence of the lesions in the subject is determined using an age-dependent model, wherein the determination uses the subject's known actual age and the one or more methylation levels, and wherein the model is generated using a reference sample of subjects with known classifications of the lesions.

31. The method of claim 30, wherein the first group of the set of CpG sites of the one or more groups corresponds to one or more genes, and wherein the one or more genes include the clusters in Table 1.

32. The method of claim 30, wherein determining the classification using the model comprises comparing the one or more methylation levels with one or more thresholds, wherein the one or more thresholds depend on the known actual age.

33. The method of claim 30, wherein the model is an age-dependent machine learning model.

34. The method of claim 33, wherein the age-dependent machine learning model comprises a plurality of sub-models, each corresponding to a different actual age.

35. The method of claim 33, wherein the set of one or more groups of CpG sites is a set of multiple groups of CpG sites, and wherein the classification for determining the presence of the lesion in the object comprises: A feature vector is generated based on the methylation levels of one or more CpG sites from the plurality of groups; and The feature vector is compared with a representative reference vector, which is determined using a set of reference vectors of measurements with the same known lesion classification.

36. The method of claim 33, wherein determining the classification of the presence of the lesion in the object comprises: The age-dependent machine learning model is used to predict the biological age of the object; Compare the biological age with the known actual age; and The classification of the lesions in the object is determined based on the comparison.

37. The method of claim 36, wherein determining the classification of the presence of the lesion in the object further comprises: Determine the difference between the biological age and the known actual age; and The difference is compared with a threshold to determine the classification of the presence of the lesion in the object.

38. The method of claim 30, wherein the same morphological classification is selected from linear, logarithmic, quadratic, and exponential.

39. The method of claim 30, wherein each group of the set of CpG sites in the one or more groups is associated with a specific tissue type, and wherein the lesion is a lesion specific to the specific tissue type.

40. The method of claim 39, wherein each group of the set of CpG sites in the one or more groups is associated with the particular tissue type based on (1) a biological pathway or (2) an epigenetic pattern or (3) an expression level greater than a threshold in the particular tissue type.

41. The method of claim 39, wherein the specific tissue type is a tissue type specific to a specific organ.

42. The method of claim 39, wherein the particular tissue type is selected from Table 2, and wherein the set of CpG sites of one or more groups is selected from the genes listed in Table 2 that are associated with the particular tissue type.

43. The method of claim 39, wherein the specific tissue type is selected from: bone marrow, brain, ovary, pancreas, liver, hypothalamus, heart, kidney, bladder, prostate, lymph nodes, breast, lung, skin, and testis.

44. The method according to any one of claims 14 to 43, wherein the lesion is a tumor.

45. The method of claim 44, wherein the lesion is a glioma.

46. ​​The method according to any one of the preceding claims, wherein the sequence readout is determined using sequencing or probe-based techniques.

47. The method of claim 46, wherein the sequencing comprises determining the methylation state by (1) treating the plurality of free DNA molecules (e.g., with bisulfite or restriction enzyme) or (2) analyzing optical or electrical signals of the plurality of free DNA molecules at locations within a window including the site.

48. The method according to any one of the preceding claims, wherein the sequence readings are paired end readings.

49. The method according to any one of claims 30 to 48, wherein determining the methylation level of the set of CpG sites of the group comprises determining an amount of methylation at the set of CpG sites that indicates the presence or absence of the methylation.

50. The method of claim 49, wherein the methylation level is the methylation density.

51. The method of claim 49, wherein the methylation level comprises the proportion of methylation states at the set of CpG sites that indicate the presence or absence of the methylation.

52. The method according to any one of the preceding claims, wherein the set of CpG sites of the one or more groups corresponds to one or more genes, and wherein the methylation level is a gene-specific methylation level.

53. A computer product comprising a non-transitory computer-readable medium storing a plurality of instructions, which, when executed, cause a computer system to perform the method as described in any one of the preceding claims.

54. A system comprising: The computer product according to claim 53; and One or more processors configured to execute instructions stored on the computer-readable medium.

55. A system comprising means for performing any of the methods described above.

56. A system comprising one or more processors configured to perform any of the methods described above.

57. A system comprising modules that respectively perform the steps of any of the methods described above.

Images